- Email: [email protected]
Interleukin 1 genetic tests provide no support for reduction of preventive dental care
ORIGINAL CONTRIBUTIONS
Interleukin 1 genetic tests provide no support for reduction of preventive dental care Scott R. Diehl, PhD; Fengshen Kuo, MS, PhD; Thomas C. Hart, DDS, PhD
A
study of dental benefits claims concluded that genetic polymorphisms in the interleukin 1 (IL-1) genes IL1A and IL1B (the “PST” genetic test) identified “high-risk” patients who benefitted significantly more than “low-risk” patients from receiving a second annual preventive visit by reducing tooth extractions.1 The authors of the study concluded that the data justify providing 2 annual preventive dental visits to only patients with diabetes, smokers, or those classified as “high risk” by their genetic test.1 They further claimed that the study also supported a different IL-1 test (the “PerioPredict” genetic test) that is currently being used to determine reimbursement for levels of preventive care by one major insurer. A recent review of clinical studies2 showed that independent reanalyses led to changes in findings and conclusions different from those of the original studies over one-third of the time. Unfortunately, independent reanalyses are rarely carried out on the data reported in published studies. This raises concern that the scientific community and the general public may be accepting findings that are of questionable validity. Large studies, such as the tooth extraction study mentioned above, are time-consuming and costly, thus are unlikely
ABSTRACT Background. It has been proposed that the PST and PerioPredict genetic tests that are based on polymorphisms in interleukin 1 (IL-1) genes identify a subset of patients who experience fewer tooth extractions if provided with 2 annual preventive visits. Economic analyses indicate rationing preventive care to only “high-risk” genotypes, smokers, patients with diabetes, or combinations of these risk factors would reduce the cost of dental care by $4.8 billion annually in the United States. Methods. Data presented in the study that claimed clinical utility for the PST and PerioPredict tests were obtained for reanalysis using logistic regression to assess whether the PST genetic test, smoking, diabetes, or number of preventive visits were risk factors for tooth extraction during a span of 16 years. Consistency of risk classification by the PST (version 1) and PerioPredict (version 2) genetic tests was evaluated in different ethnic groups from the 1000 Genomes database. Results. Multivariate analyses revealed association of tooth extraction with diabetes (P < .0001), smoking (P < .0001), and number of preventive visits (P ¼ .004), but no support for the PST genetic test (P ¼ .96) nor indication that the benefit of 2 preventive visits was affected by this genetic test (P ¼ .58). Classification of risk was highly inconsistent between the PST (version 1) and PerioPredict (version 2) genetic tests. Conclusions. Two annual preventive visits were supported as beneficial for all patients, and there was no evidence that the IL-1 PST genetic test has any effect on tooth extraction risk or influences the benefits of 2 annual preventive visits. Practical Implications. Neither IL-1 PST nor PerioPredict genetic tests are useful for rationing preventive dental care. Further research is needed to identify genetic biomarkers with robust clinical validity and clinical utility to effectively personalize the practice of dentistry. Key Words. Genetic screening; genetic test; biomarker; single nucleotide polymorphism; personalized medicine; tooth extraction; tooth loss; preventive care; clinical data reanalysis. JADA 2015:146(3):164-173 http://dx.doi.org/10.1016/j.adaj.2014.12.018
Dr. Diehl is a professor, Department of Oral Biology, Rutgers School of Dental Medicine, and professor of health informatics, Rutgers School of Health Related Professions, Rutgers Biomedical Health Sciences, MSB C636, Newark, NJ 07101, e-mail [email protected]. Address correspondence to Dr. Diehl. Dr. Kuo is a senior bioinformatics scientist, Department of Oral Biology, Rutgers School of Dental Medicine, Rutgers Biomedical Health Sciences, Newark, NJ. Dr. Hart is a professor, Department of Periodontics, and Director, Craniofacial Population Sciences Research, University of Illinois at Chicago College of Dentistry, Chicago, IL. He also is the chair of the ADA Council on Scientific Affairs, Chicago, IL, and will be the director of the ADA Foundation Dr. Anthony Volpe Research Center, Gaithersburg, MD, starting April 2. Copyright ª 2015 American Dental Association. All rights reserved.
164 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
to be replicated. Therefore, it is essential that complete data be made available for independent reanalysis to increase assurance of the validity of study conclusions.3 Data in the original article on risk factors for tooth extraction and patient stratification were insufficient to perform an independent reanalysis.1 Specifically, patients who have diabetes and/or were smokers—2 wellestablished risk factors for tooth loss—were pooled together within “high-risk groups” that also included patients who were classified as “high risk” based solely on their PST genotype test. Consequently, it was not possible to evaluate whether the PST genetic test itself had any effect on the clinical outcomes independent of smoking and/or diabetes. In addition to the PST test (based on 1 polymorphism in the IL1A gene and 1 in the IL1B gene), the original article also introduced a new genetic test, PerioPredict (that the authors called version 2) based on 4 other polymorphisms all located in the IL1B gene region. It was claimed that the PerioPredict test was equivalent for clinical purposes to its predecessor, the PST genetic test.1 Although the individual single nucleotide polymorphisms (SNPs) for the new PerioPredict genetic test were reported, the way by which the genotype data from the 4 SNPs were used to classify patients into “high-risk” versus “low-risk” groups was not presented in the published study.1 To perform an independent reanalysis of this study, we requested this information from the authors. We obtained data on tooth extraction outcomes separately by participants’ smoking, diabetes, PST genotype risk group, and number of preventive visits (W.V. Giannobile, DDS, DMSc, e-mail communication, December 2013). Patient age, race, or sex was not disclosed to us so these demographic measures could not be included as covariates in our reanalysis. To our knowledge, data on patients’ oral hygiene, numbers of teeth present, caries, or restorations at the start of the study period were neither obtained directly from the participants under the original study protocol nor available in the insurance database of dental procedures. In addition, the algorithm by which the 4 IL1B region SNPs were used to classify patients as “high risk” versus “low risk” for the PerioPredict test was also shared (W.V. Giannobile, DDS, DMSc, e-mail communication, February 2014). In reviewing the analyses performed in the original article based on insurance claims,1 we noted a number of potential problems and omissions of critical statistical tests needed to support the stated conclusions. As noted above, smoking and diabetes were confounded with potential effects of the IL-1 PST genetic test. Most important, although the authors claimed that “Interactions of risk status and frequency of preventive visits on tooth loss were evident .” they did not actually report results of any interaction tests.1 In fact, there was not even any direct evaluation of whether the PST genetic test had any effect at all on risk of undergoing tooth extraction (that is, whether
or not it was a “main effect”). Furthermore, no direct evaluation of whether the PST genetic test classifies individual patients as “high risk” or “low risk,” consistent with how the PerioPredict genetic test classifies the same patients, was presented. A figure in the appendix of the published study1showed comparisons of the percentage of patients with tooth loss events for “low-risk” versus “highrisk” groups and the authors of the study concluded that their version 2 (PerioPredict) test “gave results comparable with those in version 1 relative to differences between 1 and 2 preventive visits.” However, the “high-risk” group presented in their figure confounded patients who are classified as “high risk” solely because of their PST or PerioPredict genetic tests with other patients who were “high risk” because they have diabetes or were smokers. Furthermore, only group frequencies were presented in the figure and there was no way of knowing if individual patients were consistently classified as “low risk” or “high risk,” or may change their risk classification depending on which test is used. There is great potential for advances in genomics to expand our knowledge of oral disease etiology and to improve patient care through personalized approaches to diagnosis, treatment, and prevention of oral and dental diseases. However, it is essential that these powerful strategies are supported by objective and independent assessment of robust clinical data free from conflict of interest.4 Implementation of patient stratification for preventive dental care based on the IL-1 PerioPredict genetic test carries significant implications for the standard of care in dentistry. Therefore, it is essential that the data from clinical studies such as this be available for independent reanalyses. The purpose of our study was to conduct a reanalysis of data provided by the authors of the published review of dental benefits claims and discuss why our findings lead to very different conclusions. METHODS
Using only the data presented in the published study of IL1A/IL1B genotypes and tooth extractions,1 it was not possible to reanalyze the authors’ most important conclusion that an interaction exists between the PST genetic test and number of preventive visits. We obtained from W.V. Giannobile, DDS, DMSc (e-mail communication, December 2013) the numbers of participants with 1 versus 2 annual preventive visits and the ABBREVIATION KEY. ASW: African Americans from the Southwest. CEU: Caucasians from Utah. CHB: Han Chinese from Beijing. FCGR: Fc fragment of IgG, low affinity IIa or IIb, receptor. HLA: Human leukocyte antigen. IL-1: Interleukin 1. LTF: Lactotransferrin. PUR: Puerto Ricans. SNPs: Single nucleotide polymorphisms. TLR: Toll-like receptor. TNF: Tumor necrosis factor. VDR: Vitamin D receptor.
JADA 146(3) http://jada.ada.org
March 2015 165
ORIGINAL CONTRIBUTIONS
percentage in each visit group who had 1 or more tooth extractions over a period of 16 years for the following risk groups: - no risk factors; - “high-risk” IL-1 PST genetic test only; - diabetes only; - smoking only; - diabetes and smoking; - diabetes and “high-risk” IL-1 PST genetic test; - smoking and “high-risk” IL-1 PST genetic test; - all 3 risk factors (diabetes, smoking, and “high-risk” IL-1 PST genetic test) as reported in eTable 1 (available online at the end of this article). The primary outcome of the originally published study was the binary classification of each patient as those who had no tooth extraction over a follow-up period of 16 years or those who had 1 or more tooth extractions (ADA codes D7010, D7140, D7210) during the same period.1 Therefore, we applied established methods of categorical data analysis to evaluate the hypothesis that any of the risk factors listed above interact with the number of preventive visits in altering this clinical outcome. Unlike the original study in which participants were classified into a single “high-risk” group if they had diabetes, were smokers, or had been identified as having the “high-risk” IL-1 PST genotypes (thus, confounding effects of these potential risk factors), we first evaluated each risk factor separately by performing univariate logistic regression. We tested the hypotheses as to whether diabetes, smoking, number of preventive visits, or the IL-1 PST genetic test were independently associated as main effects with the tooth extraction clinical outcome. Next, we conducted multivariate logistic regression that included all 4 main effects simultaneously in the model to determine if their effects act independently. Lastly, we tested all 2-way interactions of risk factors including interaction with number of preventive visits with the IL-1 PST genetic test that was used for all of the statistical analyses reported in the dental insurance claims study. Interactions were tested in models that contained all main effects and 1 interaction term at a time in the model. Three-way interactions were not tested as sample size (especially patients with diabetes and smokers, as shown in the results) was too small to have any reasonable power. All analyses in this study were performed using JMP Genomics version 6.1 and SAS version 9.4 computer programs (SAS Institute, Cary, NC). The traditional P value of < .05 was used to assess whether an outcome was statistically significant; when we obtained much smaller P values, this was considered when interpreting the findings. We next evaluated subgroups of patients to see whether interaction between the IL-1 PST genetic test and number of preventive visits might be found within 1 of the subgroups even if not present in the sample as a whole. For example, in patients who neither had diabetes nor were smokers, we tested whether being classified as
166 JADA 146(3) http://jada.ada.org
March 2015
“high risk” or “low risk” by the IL-1 PST genetic test interacts with the benefits of a second annual preventive visit. Similar subgroup analyses were performed for patients who had diabetes, were smokers, or both had diabetes and were smokers. All tests of statistical significance reported in Giannobile and colleagues’1 study were based on what they called “version 1” of their IL-1 composite genotype test (also known as the “PST” test, abbreviated here as “PST genetic test”). They also introduced a different IL-1 genetic test that they called “version 2” (also known as the “PerioPredict” test, abbreviated here as the “PerioPredict genetic test”) based on 4 different SNPs. However, as noted above, the information presented in their publication was insufficient to classify participants as “high risk” versus “low risk” for the IL-1 PST (version 1) or the PerioPredict (version 2) genetic tests in independent samples. We requested and obtained this information from W.V. Giannobile, DDS, DMSc (e-mail communication, February 2014) as reported in eTables 2 and 3 (available online at the end of this article). Frequencies of “high-risk” versus “low-risk” genotypes were compared for the PST and PerioPredict genetic tests in 61 African Americans recruited in the Southwest United States (ASW), 55 Puerto Ricans (PUR) recruited in Puerto Rico, 55 Caucasians from Utah (CEU), and 97 Han Chinese recruited in Beijing (CHB) using data from the 1000 Genomes database Phase 1 release.5 The objectives were to measure variation in the frequency of “high-risk” participants in different populations and ethnic groups for each test and to determine how consistently the 2 different tests classified participants as high versus low risk within each group. RESULTS
We first evaluated all participants in a combined analysis to determine which risk factors showed evidence of association with the clinical outcome, tooth extraction. Odds ratios shown in Table 1 for univariate analyses revealed significant increased risk of undergoing the tooth extraction outcome with the main effects: diabetes (P < .0001), smoking (P < .0001), and decreased risk associated with 2 annual preventive visits (P ¼ .0004). In contrast, there was no evidence of association of the IL-1 PST genetic test with the tooth extraction outcome (P ¼ .8604). Multivariate analyses, in which all main effect variables are included in the model simultaneously, were performed to see if these risk factors act independently of each other. Essentially identical results were obtained, indicating that effects of diabetes, smoking, and number of preventive visits were each independently associated with the tooth extraction outcome, while the IL-1 PST genetic test still showed no evidence of significance (P ¼ .9568). Next, we tested if any risk factors gave different results depending on whether patients received 1 or 2 annual
ORIGINAL CONTRIBUTIONS
preventive visits or if other combinations of variables showed evidence of interaction. Interaction term parameters, standard errors, and P values are shown in Table 1. The results indicate no evidence that the benefits of 2 preventive visits differed depending on patients’ other risk factors. This indicates no support for interaction of number of preventive visits with diabetes (P ¼ .8074), smoking (P ¼ .4236), or the IL-1 PST genetic test (P ¼ .5788). No combination of other pairs of variables showed evidence of interaction. Results of stratified analyses were fully consistent with these findings. Three-quarters of patients in the study neither had diabetes nor were smokers. These included 1,186 participants with 1 annual preventive visit and 2,674 participants with 2 visits (Figure 1). In this subgroup, patients who received 2 annual preventive visits had lower levels of tooth extraction over a period of 16 years (P ¼ .008). The IL-1 PST genetic test had no effect on overall risk of undergoing tooth extraction (P ¼ .589) in this group of participants who neither have diabetes nor were smokers. Furthermore, the benefit of having 2 annual preventive visits was not influenced by whether a patient was classified as either “high risk” or “low risk” by the IL-1 PST genetic test (that is, there was no evidence of an interaction between visit number and the genetic test, P ¼ .475). These results of Figure 1 for patients who were nonsmokers and did not have diabetes are presented in additional detail in the first two rows of Table 2. Other subgroups shown in Table 2 include patients who had diabetes but were not smokers, smokers who did not have diabetes, and participants who both had diabetes and were smokers. There was no evidence for an effect of the PST genetic test on the tooth extraction clinical outcome in any of these subgroups (P values ¼ .814, .606, and .258, respectively) nor any support for the hypothesis that the benefit of 2 annual preventive visits in reducing tooth extraction is influenced by this genotype test (interaction P values ¼ .251, .568, and .836, respectively). It should be noted that the observed relationships between the IL-1 PST genetic test and tooth loss is highly inconsistent among patients with diabetes and smokers. For example, a higher percentage of patients who had diabetes but were nonsmokers had tooth extractions when they received 2 annual preventive visits and were “high risk” according to the PST genetic test (24.1%) versus having only 1 visit (19.0%). However, for patients who were smokers but did not have diabetes, the trend went in the opposite direction for the PST genetic test “high-risk” group (21.5% with 2 visits versus 31.0% with 1 visit). However, none of these differences approached statistical significance and were likely due to sampling error expected because of the small number of participants in these groups. Odds ratios and 95% confidence intervals (CI) for each risk factor are shown in Figure 2. No interaction term was included in this model because none was
TABLE 1
Logistic regression analysis of potential risk factors for tooth loss over 16 years. UNIVARIATE ANALYSES* P Value
Main Effects
Odds Ratio
95% Confidence Interval
IL-1† PST genetic test “high risk”
1.01
0.87-1.18
.8604
Diabetes
1.71
1.34-2.18
< .0001
Smoking
1.85
1.56-2.19
< .0001
2 visits
0.76
0.65-0.88
.0004
MULTIVARIATE ANALYSES* P Value
Main Effects
Odds Ratio
95% Confidence Interval
IL-1 PST genetic test “high risk”
1.00
0.86-1.17
Diabetes
1.66
1.28-2.10
.0001
Smoking
1.84
1.55-2.18
< .0001
.9568
0.76
0.65-0.88
.0005
Interactions‡
Estimate
Standard Error
P Value
Diabetes X visit
0.02
0.06
.8074
Smoking X visit
0.04
0.05
.4236
IL-1 PST X visit
0.02
0.04
.5788
IL-1 PST X diabetes
0.05
0.06
.4761
IL-1 PST X smoking
0.04
0.05
.3680
Diabetes X smoking
0.01
0.07
.8755
2 visits
* Univariate models for the binary outcome of none versus 1 or more tooth extractions over a period of 16 years, as in the original publication,1 were run separately for each main effect shown with only that variable in the model. Multivariate main effects’ results are based on a single model with all variables present in the model but no interactions. † IL-1: Interleukin 1. ‡ Interaction models included all 4 main effects but only 1 interaction term in each model, run separately for each of the 6 pairwise interactions shown.
statistically significant. Having 2 annual preventive visits was protective versus 1 visit, and both smoking and diabetes significantly increased risk of undergoing tooth extraction. After accounting for the effects of diabetes and smoking, patients classified as “high risk” by the PST genetic test in the published study1 did not exhibit any elevation of risk whatsoever with an odds ratio of 1.00 (95% CI 0.86-1.17). Although all tests of statistical significance reported in the published study1 were based on the PST genetic test, Interleukin Genetics has launched commercial testing for dental insurance coverage using the PerioPredict genetic test. The PerioPredict genetic test is based on 4 entirely different genetic polymorphisms than the PST genetic test but the authors claimed “either version of the IL-1 test to classify low- and high-risk patients provides the same results with regard to the relationship of the
JADA 146(3) http://jada.ada.org
March 2015 167
% OF PATIENTS WITH TOOTH EXTRACTION
ORIGINAL CONTRIBUTIONS
18%
16%
14%
12%
10% 1 Visit IL-1 PST “High-Risk” Genotypes
2 Visits IL-1 PST "Low-Risk" Genotypes
Figure 1. Percentages (standard error) of patients who had neither diabetes nor were smokers who experienced 1 or more tooth extractions over 16 years, classified as “high risk” or “low risk” by the interleukin 1 (IL-1) PST genetic test and received either 1 or 2 annual preventive visits. Having 2 visits was protective (P ¼ .008), but the IL-1 PST genotypes had no effect on risk (P ¼ .589); there is no support for interaction between number of visits and the genotype test (P ¼ .475).
number of preventive visits and tooth loss.”1 If the 2 tests actually are equivalent, they should assign the same level of risk (that is, “high risk” or “low risk”) to each patient. To evaluate test consistency, participants from 4 ethnic groups included in the 1000 Genomes database were classified as “high risk” versus “low risk” using the risk definitions of the PST and PerioPredict genetic tests (eTables 2 and 3; available online at the end of this article). As shown in Figure 3, the percentage of patients classified as “high risk” varies widely among different ethnic groups even for the same test. Furthermore, the PST and PerioPredict tests classify as “high risk” different percentages of the same population. For example, although only 11% of the ASW group were classified as “high risk” by the PST genetic test, the percentage of “high-risk” patients in this population increased more than fivefold to 61% when classified using the PerioPredict genetic test. In contrast, CEU participants decreased from 34% classified as “high risk” by the PST test to 18% “high risk” according to the PerioPredict test. Some populations, such as the PUR participants in the study, did not show any change in frequency of “high risk” participants with 25% reported as “high risk” for both the PST and PerioPredict genetic tests (Figure 3). However, classification of individual patients was inconsistent between the 2 tests. A substantial number of participants changed from “high risk” to “low risk,” or vice versa, depending on which test was used. As shown
168 JADA 146(3) http://jada.ada.org
March 2015
in Figure 4, 33% of the PUR participants had their risk classification change either from high to low risk or from low to high risk depending on which version of these IL-1 tests was used. Over one-half of ASW participants (56%), 28% of CEU, and 13% of CHB participants were assigned different risk statuses by the PST and PerioPredict genetic tests, demonstrating that the tests were not the same and were highly inconsistent. DISCUSSION
This reanalysis of a published study using data from insurance claims found no evidence for association of the IL-1 PST genetic test with risk of undergoing tooth extraction. The well-established risks associated with diabetes6 and smoking7 were confirmed, whereas having 2 annual preventive visits versus only 1 visit was associated with reducing the odds of patients having 1 or more tooth extractions over a period of 16 years by approximately 25%. Most of the comparisons in the previously reported analysis involved contrasting participants with no risk factors versus a pool of patients who had diabetes or were smokers and/or were identified as “high risk” by the IL-1 PST genetic test, confounding these variables as a single putative “highrisk” group. Our reanalysis showed that when the PST genetic test was evaluated on its own or in a multivariate model that included terms for these other risk factors (Table 1), it has no effect at all on tooth extraction risk in this data set (OR ¼ 1.0). In the data reanalysis, for most patients in the study who neither had diabetes nor smoked, being identified by the PST genetic test as “high risk” had no significant effect on risk of undergoing tooth extraction (Figure 1, P ¼ .59) whereas 2 annual preventive visits was consistently beneficial for patients (P ¼ .008), regardless of their PST genetic test status (P ¼ .47). The claim made in the published study by Giannobile and colleagues1 that potentially could have the greatest impact on the standard of care in dentistry was that only those patients who had diabetes, were smokers, or were classified as “high risk” by the IL-1 PST genetic test benefit when provided with 2 annual preventive visits by reducing risk of undergoing tooth extraction. Most
ORIGINAL CONTRIBUTIONS
TABLE 2
Stratified analyses of subgroups of patients by risk factors. RISK FACTORS
1 VISIT
Diabetes
Smoking
Interleukin 1 PST Genetic Test
No
No
Low risk
No
No
High risk
Yes
No
Low risk
No. of Participants
2 VISITS Tooth Loss (%)
DIFFERENCE (%)
Tooth Loss (%)
No. of Participants
732
16.4
1,686
13.8
2.6
454
17.0
988
12.7
4.3
67
25.4
112
18.8
6.6
Yes
No
High risk
42
19.0
79
24.1
5.0
No
Yes
Low risk
158
26.6
386
21.2
5.3
No
Yes
High risk
100
31.0
223
21.5
9.5
Yes
Yes
Low risk
19
36.8
33
27.3
9.6
Yes
Yes
High risk
12
50.0
26
34.6
15.4
P VALUES Visit
PST Genetic Test
PST Genetic Test X Visit
.008
.589
.475
.682
.814
.251
.028
.606
.568
.358
.258
.836
ODDS RATIO
patients in the study1 were classified as “low 6 risk” by the PST genetic on cti test and it was claimed tra x E 5 they would not benefit oth To g from 2 annual prevenoin erg tive visits. If this were d n 4 fU true, it would mean that ko s i R an interaction exists ed as e between the IL-1 PST r 3 Inc genetic test and number of preventive visits. 2 However, no interaction Protective No Effect tests were presented in the published study.1 In 1 our reanalysis, results of Diabetes this important interac2 Visits 1 Visit Diabetes Smoking IL-1 PST & Smoking tion test have been pre0 sented both for the RISK FACTORS FOR TOOTH EXTRACTION entire data set and in stratified analyses of subgroups. Of particular Figure 2. Odds ratios and 95% confidence intervals for each risk factor main effect in the multivariate model. Two interest are the results annual preventive visits reduced the risk of experiencing tooth loss compared with 1 annual visit. Diabetes, smoking, for the subgroup of pa- and both risk factors together significantly increased risk. The interleukin 1 (IL-1) variable showed that the “high-risk” tients who were neither IL-1 PST genetic test had no effect on risk of experiencing tooth loss in this study over a period of 16 years. smokers nor had diabetes. In no instance was there any evidence or even a 988 with 2 visits). Our finding of neither a main effect nor weak suggestion of support for this interaction. Therean interaction for this genetic test cannot be attributed to fore, our re-evaluation of this data set, using widely lack of power. In this sample, if the test had relative risk accepted and standardized methods for categorical data of 1.35 or greater, it would be expected to yield a P < .05 analysis, leads to the conclusion that a second annual over 80% of the time (eTable 4; available online at the end preventive visit was beneficial for all participants, of this article). The diabetes and smoking risk factors regardless of their IL-1 PST genetic test risk classification revealed highly significant association with tooth or whether they had diabetes or were smokers (Figure 2). extraction (Table 1) even though these groups had much There was a substantial number of patients who did smaller sample sizes (140 patients with diabetes with 1 not have diabetes and were nonsmokers with “low-risk” visit and 250 with 2 visits; 289 smokers with 1 visit, PST genotypes (732 patients with 1 visit and 1,686 with 668 with 2 visits). 2 visits) available for comparison with patients having The findings from our reanalysis are different from “high-risk” PST genotypes (454 patients with 1 visit and those of the published study. Reanalysis occurs
JADA 146(3) http://jada.ada.org
March 2015 169
ORIGINAL CONTRIBUTIONS
% OF SUBJECTS INCONSISTENT
% OF CLASSIFIED AS "HIGH RISK"
shared access to clinical study data,3,8 the focus of a project by the 70% Institute of Medicine, 61% 60% “Strategies for Responsible Sharing of Clinical 50% Trial Data.”9 Our reanalysis shows that the 40% 34% benefits of data sharing 30% extend to clinical 25% 25% studies in dentistry. 18% 20% We have also shown 11% 11% that the PST and 10% 4% PerioPredict genetic tests are different. They 0% PST (Version 1) PerioPredict (Version 2) classify risk for the INCONSISTENCY ACROSS TESTS same patient inconsistently a large portion of ASW CHB PUR CEU the time, and both tests vary substantially in the frequency of patients Figure 3. Percentages of participants classified as “high risk” by the interleukin 1 (IL-1) PST (version 1) and classified as “high risk” PerioPredict (version 2) genetic tests in African Americans from the Southwest (ASW), Puerto Ricans (PUR), across different races Caucasians from Utah (CEU), and Han Chinese from Beijing (CHB). Risk assignments differ substantially both and ethnic groups. across the different ethnic groups and between the PST and PerioPredict tests. Because of this, there are essentially no clinical data on tooth extraction frequencies 60% using the PerioPredict 56% genetic test. This raises special concern because 50% the PerioPredict genetic test is the version that is 40% now being marketed for 33% clinical practice and 28% dental insurance 30% purposes. We suggest there are 20% at least 2 reasons why 13% the IL-1 PST genetic test 10% has no effect on risk of undergoing tooth extraction. IL-1 genetic 0% ETHNIC GROUPS polymorphisms have been associated with ASW PUR CEU CHB risk of developing periodontitis at only weak to moderate strength, and periodontitis is Figure 4. Percentages of individual participants whose risks are classified differently by the interleukin 1 (IL-1) PST only 1 of many causes (version 1) and PerioPredict (version 2) genetic tests in African Americans from the Southwest (ASW), Puerto Ricans (PUR), Caucasians from Utah (CEU), and Han Chinese from Beijing (CHB). of tooth extraction in middle-aged adults. By infrequently in medical and dental research but a recent contrast, 2 underlying assumptions of the originally review of 37 reanalyses of clinical studies found that 35% published study were that the PST genetic test identifies of these led to conclusions different from the original subjects who have substantially increased risk of periodontitis, and that periodontitis, in turn, is the major report.2,3 This shows the merit in providing broad,
170 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
IL-1 PST
Tooth Extraction
Periodontitis
A
Smoking
Gender Diabetes Host Genes: IL1 IL2 IL6 IL10 FCGRs HLA LTF TLRs TNFs VDR and many other genes
Age
Oral Microbiome
Alcohol
Caries Tooth Extraction
Periodontitis
Smoking
Oral Microbiome
Failed Restorations
Root Canal Host Genes Trauma
B Figure 5. A. Simple relationships between interleukin 1 (IL-1) PST genotypes and risk of developing periodontitis and tooth extraction. B. More realistic model of the complex etiologies with multiple risk factors for periodontitis and multiple causes of tooth extraction illustrating that periodontitis is only 1 of many risk factors for this outcome. FCGR: Fc fragment of IgG, low affinity IIa or IIb, receptor. HLA: Human leukocyte antigen. LTF: Lactotransferrin. TLR: Toll-like receptor. TNF: Tumor necrosis factor. VDR: Vitamin D Receptor.
cause of tooth extraction in middle-aged adults (Figure 5A). In fact, the situation is far more complex as illustrated in Figure 5B. Association studies of the PST “high risk” genotypes (and numerous other IL-1 “composite genotypes” and individual SNPs reported in the literature over the years) with chronic periodontitis indicates that this inherited variation has at most only a weak-tomoderate effect on disease risk.10 Even in studies with highly-significant P values, many periodontitis cases do not have the “high-risk” genotype and many health controls do have this putative “high-risk” genotype. For example, among patients who did not have diabetes and were nonsmokers, 30% of chronic periodontitis cases and 20% of controls were classified as “high risk” by the IL-1 PerioPredict genetic test as reported in a recent study (Table 5 in Wu and colleagues11 where the PerioPredict genetic test is labeled “Pattern 3”). Even though the association is statistically significant (P ¼ .005), no clinically useful information is provided by the test. Furthermore, none of the genomewide association studies of periodontitis have revealed any evidence that polymorphisms in IL-1 genes are ranked anywhere among the best-supported SNPs for disease susceptibility.12-15 Thus, evidence indicates that inherited genetic variation in other genes and chromosome regions accounts for most of the large heritable component of periodontitis (Figure 5B). Furthermore, environmental
factors also play an important role in disease susceptibility.16 It is well established that tooth loss is related to age, sex, percentage of teeth with restorations, alcohol, smoking, diabetes, and number of teeth present at the start of a study.17 In addition, as we have shown here, preventive dental care also protects against tooth loss. Because periodontitis is 1 cause, but only 1 of many causes of tooth extraction (Figure 5B), it is hardly surprising that the IL-1 PST or PerioPredict genetic tests for genotype had no strong predictive association with tooth extraction as our reanalysis indicates. The data provided to us from the authors of the published study were limited to the dichotomized outcome of patients having no tooth extractions versus having 1 or more tooth extractions over a period of 16 years. Although only roughly 1 in 5 patients lost any teeth, it is possible that the small percentage of patients who lost large numbers of teeth over the 16 years might have an especially strong association with number of preventive visits or with 1 of the other risk factors considered in the study, but that possibility could not be considered in our reanalysis. SNP-based genetic testing for common complex diseases such diabetes, mental illness, cardiovascular disease, and obesity is currently of limited clinical utility, as these tests are generally no better at predicting disease risk than existing blood tests and family history.18,19 Concerns have been raised by a number of advisory
JADA 146(3) http://jada.ada.org
March 2015 171
ORIGINAL CONTRIBUTIONS
groups including the US Government Accountability Office20; the Secretary’s Advisory Committee on Genetic Testing21; the Evaluation of Genomic Applications in Practice and Prevention Working Group22; and the National Human Genome Research Institute23 regarding the limited clinical utility of SNP-based genetic testing for common, complex diseases. These bodies have noted a lack of regulation and particular concern that demonstration of clinical validity and clinical utility of such tests are not required before tests are marketed. Studies need independent replication and demonstration of clinical validity and utility as well as training of health professionals in understanding the meaning of test results so that they can explain the results clearly at a level appropriate for their patients. In Mendelian diseases such as cystic fibrosis, amelogenesis imperfecta, and some forms of tooth agenesis, mutation of a single gene causes the disease.24 In contrast, it is now widely accepted among human geneticists that heritability for common complex diseases such as type 2 diabetes, rheumatoid arthritis, and myocardial infarction, with rare exceptions, is due to the accumulated effects of genetic variation at hundreds or thousands of genes. Each variant has a weak to moderate effect, similar in magnitude to that of the weak to moderate effect of IL-1 SNPs on risk of developing chronic periodontitis.25 Because so many DNA variants are involved in risk of developing complex disease, and because there are often important environmental and epigenetic effects too, even when individual SNPs are disease-associated with strong statistical significance, they have little ability to predict disease risk in patients. For example, the SNP rs10993994 in the MSMB gene is associated with risk of developing prostate cancer (P value of 8.7 10-29).26 This gene is synthesized by the epithelial cells of the prostate gland and has effects on cell proliferation and apoptosis, making it a good candidate gene for prostate cancer. However, the “high-risk” C allele at this strongly associated SNP has a frequency of 38% in controls and 46% in cases. Knowing which of these alleles a patient has does not provide clinically actionable information either for the patient or for their doctor. Because the SNP is only 1 of many etiologic risk factors, and accounts for only a small amount of overall disease risk, people often develop disease with or without this specific SNP. Thus, using this individual SNP as a “stand-alone” test is not clinically useful for disease risk prediction. In spite of the great challenges of complex diseases, progress is being steadily made. Genomic technologies are continuing to rapidly advance, with high-throughput sequencing of DNA and RNA, proteomic, and epigenetic profiling providing increasing coverage at lower costs. Importantly, bioinformatics and complex statistical strategies and approaches, often involving high-performance computing, are also enhancing clinical
172 JADA 146(3) http://jada.ada.org
March 2015
researchers’ capabilities. Success will also clearly require assembly of much larger numbers of patients and controls than employed in most dental genetic studies conducted to date. One recent example that may suggest a path ahead is a study of type 1 diabetes that included 4,574 cases and a genetic score based on summation of SNPs at 40 non-HLA (human leukocyte antigen) genes that have been identified from replicated large genomewide studies.27 Use of the genetic score significantly improved disease prediction beyond that provided by HLA alone. That study27 and others using genetic score approaches including both modifiable and nonmodifiable risk factors28 are showing promise, but even with dozens of SNPs, these approaches are still not ready for use in the clinical setting for disease prediction. Finally, it should be noted that even when SNPs are not useful for disease prediction, genetic studies of complex diseases have already provided many important insights into previously unknown but etiologically important biological pathways that point toward novel approaches for disease prevention and therapeutic targets.29 CONCLUSIONS
Our reanalysis of a tooth extraction study1 that used insurance claims spanning 16 years has led to conclusions different from those reported in the original article. The data provide support for 2 annual preventive visits as beneficial for all patient groups and confirm the well-established effects of smoking and diabetes as risks for tooth extraction in adults. In contrast, the IL-1 PST genetic test has no significant relationship to tooth extraction risk, and there is no evidence that the benefits of having 2 annual preventive visits differ between “high-risk” and “low-risk” genotype groups. The IL-1 PST and PerioPredict genetic tests are not equivalent, as they classify patients’ risks differently, and neither test has any evidence supporting their use for patient stratification or any other aspect of dental care. n SUPPLEMENTAL DATA
Supplemental data can be found at: http://dx.doi.org/ 10.1016/j.adaj.2014.12.018. Disclosure. None of the authors reported any disclosures. The authors thank Olga Korczeniewska for assistance with initial data analyses and graphics, and Rushi Patel for help preparing tables and figures. 1. Giannobile WV, Braun TM, Caplis AK, Doucette-Stamm L, Duff GW, Kornman KS. Patient stratification for preventive care in dentistry. J Dent Res. 2013;92(8):694-701. Available at: http://jdr.sagepub.com/content/92/8/ 694.full. Accessed January 20, 2015. 2. Ebrahim S, Sohani ZN, Montoya L, et al. Reanalyses of randomized clinical trial data. JAMA. 2014;312(10):1024-1032. 3. Krumholz HM, Peterson ED. Open access to clinical trials data. JAMA. 2014;312(10):1002-1003. 4. Offit K. Personalized medicine: new genomics, old lessons. Hum Genet. 2011;130(1):3-14.
ORIGINAL CONTRIBUTIONS
5. 1000 Genomes Project Consortium; Abecasis GR, Auton A, Brooks LD, et al. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491(7422):56-65. 6. Kapp JM, Boren SA, Yun S, LeMaster J. Diabetes and tooth loss in a national sample of dentate adults reporting annual dental visits. Prev Chronic Dis. 2007;4(3):A59. 7. Hanioka T, Ojima M, Tanaka K, Matsuo K, Sato F, Tanaka H. Causal assessment of smoking and tooth loss: a systematic review of observational studies. BMC Public Health. 2011;11:221. 8. Tucker ME. How should clinical trial data be shared? BMJ. 2013;347: f4465. 9. Institute of Medicine. Sharing Clinical Trial Data: Maximizing Benefits, Minimizing Risk. Washington, DC: The National Academies Press; 2015. 10. Diehl SR, Chou CH, Kuo F, Huang C-Y, Korczeniewska O. Genetic susceptibility to periodontal disease. In: Newman MG, Takei HH, Klokkevold PR, Carranza FA, eds. Carranza’s Clinical Periodontology. 12th ed. St. Louis, MO: Elsevier Saunders; 2014:101-115. 11. Wu X, Offenbacher S, López NJ, et al. Association of interleukin-1 gene variations with moderate to severe chronic periodontitis in multiple ethnicities. J Periodontal Res. 2015;50(1):52-61. 12. Schaefer AS, Richter GM, Nothnagel M, et al. A genome-wide association study identifies GLT6D1 as a susceptibility locus for periodontitis. Hum Mol Genet. 2010;19(3):553-562. 13. Divaris K, Monda KL, North KE, et al. Exploring the genetic basis of chronic periodontitis: a genome-wide association study. Hum Mol Genet. 2013;22(11):2312-2324. 14. Teumer A, Holtfreter B, Völker U, et al. Genome-wide association study of chronic periodontitis in a German population. J Clin Periodontol. 2013;40(11):977-985. 15. Rhodin K, Divaris K, North KE, et al. Chronic periodontitis genomewide association studies: gene-centric and gene set enrichment analyses. J Dent Res. 2014;93(9):882-890. 16. Genco RJ, Borgnakke WS. Risk factors for periodontal disease. Periodontol 2000. 2013;62(1):59-94. 17. Copeland LB, Krall EA, Brown LJ, Garcia RI, Streckfus CF. Predictors of tooth loss in two US adult populations. J Public Health Dent. 2004;64(1): 31-37.
18. Ginsburg GS, Shah SH, McCarthy JJ. Taking cardiovascular genetic association studies to the next level. J Am Coll Cardiol. 2007;50(10):930-932. 19. Wray NR, Yang J, Hayes BJ, Price AL, Goddard ME, Visscher PM. Pitfalls of predicting complex traits from SNPs. Nat Rev Genet. 2013;14(7):507-515. 20. Kuntz G. Direct-to-consumer genetic tests: misleading test results are further complicated by deceptive marketing and other questionable practices. Testimony before the subcommittee on oversight and investigations, committee on energy and commerce, House of Representatives; 2010. Washington, DC: US Government Accountability Office Publication GAO-10-847T. 21. Secretary’s Advisory Committee on Genetic Testing, National Institutes of Health. Enhancing the oversight of genetic tests: recommendations of the SACGT; 2000. Available at: http://osp.od.nih.gov/sites/default/files/ resources/oversight_report.pdf. Accessed February 3, 2015. 22. Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group. Recommendations from the EGAPP Working Group: does genomic profiling to assess type 2 diabetes improve health outcomes? Genet Med. 2013;15(8):612-617. 23. National Human Genome Research Institute, National Institutes of Health. Regulation of genetic tests. Available at: www.genome.gov/1 0002335. Updated September 2, 2014. Accessed November 6, 2014. 24. US National Library of Medicine, National Institutes of Health. Genetics Home Reference; 2014. Available at: http://ghr.nlm.nih.gov/. Accessed November 6, 2014. 25. Ripke S, O’Dushlaine C, Chambert K, et al. Genome-wide association analysis identifies 13 new risk loci for schizophrenia. Nat Genet. 2013;45(10): 1150-1159. 26. Liu H, Wang B, Han C. Meta-analysis of genome-wide and replication association studies on prostate cancer. Prostate. 2011;71(2):209-224. 27. Winkler C, Krumsiek J, Buettner F, et al. Feature ranking of type 1 diabetes susceptibility genes improves prediction of type 1 diabetes. Diabetologia. 2014;57(12):2521-2529. 28. Langenberg C, Sharp SJ, Franks PW, et al. Gene-lifestyle interaction and type 2 diabetes: the EPIC interact case-cohort study. PLoS Med. 2014; 11(5):e1001647. 29. Diogo D, Okada Y, Plenge RM. Genome-wide association studies to advance our understanding of critical cell types and pathways in rheumatoid arthritis: recent findings and challenges. Curr Opin Rheumatol. 2014;26(1):85-92.
JADA 146(3) http://jada.ada.org
March 2015 173
ORIGINAL CONTRIBUTIONS
eTABLE 1
Participant counts and frequency of tooth extraction (loss) by risk factor group and number of preventive visits.* NO. OF RISK FACTORS
RISK FACTOR
1 VISIT No. of Participants
2 VISITS Tooth Loss (%)
No. of Participants
TOTAL Tooth Loss (%)
0
None
732
16.4
1,686
13.8
2,418
1
Diabetic only
67
25.4
112
18.8
179
1
Smoking only
158
26.6
386
21.2
544
1
IL-1† PST
454
17.0
988
12.7
1,442
2
Diabetic and smoking
19
36.8
33
27.3
52
2
Diabetic and IL-1 PST
42
19.0
79
24.1
121
2
Smoking and IL-1 PST
100
31.0
223
21.5
323
3
Diabetic, smoking, and IL-1 PST
12
50.0
26
34.6
39
TOTAL
All Risk Factor Groups
1,584
19.4
3,533
15.4
5,119
* Data provided by W.V. Giannobile, DDS, DMSc (e-mail communication, December 2013). † IL-1: Interleukin 1.
173.e1 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
eTABLE 2
Classification of “high risk” and “low risk” by interleukin 1 PST (version 1) genetic test† GENE
IL1A
IL1B
SNP LOCUS‡
4845
3954
ALLELES
G/T
C/T
rs NUMBER
17561
1143634
Genotypes
T/*
§
TEST RESULT
T/*
High risk
T/*
C/C
Low risk
G/G
T/*
Low risk
G/G
C/C
Low risk
† Data provided by W.V. Giannobile, DDS, DMSc (e-mail communication, February 2014). ‡ SNP: Single nucleotide polymorphism. § Asterisks denote either allele at this position.
JADA 146(3) http://jada.ada.org
March 2015 173.e2
ORIGINAL CONTRIBUTIONS
eTABLE 3
Classification of “high risk” and “low risk” by interleukin 1 PerioPredict (version 2) genetic test.†‡ GENE
IL1B
IL1B
IL1B
IL1B
SNP LOCUS§
-511
-1464
-3737
3877
ALLELES
TEST RESULT
C/T
G/C
C/T
G/A
rs NUMBER
16944
1143623
4848306
1143633
Genotypes
C/C
G/G
C/C
*/*
C/C
G/G
T/T
G/G
High risk
C/T
G/C
C/C
*/*
High risk
C/T
G/G
C/C
*/*
High risk
C/T
G/G
C/T
G/G
High risk
T/T
G/*
C/C
*/*
High risk
C/C
C/*
*/*
*/*
Low risk
C/C
G/G
C/T
*/*
Low risk
C/C
G/G
T/T
A/*
Low risk
C/T
C/C
*/*
*/*
Low risk
C/T
G/C
T/*
*/*
Low risk
C/T
G/G
C/T
A/*
Low risk
C/T
G/G
T/T
*/*
Low risk
T/T
C/C
*/*
*/*
Low risk
T/T
G/*
T/*
*/*
Low risk
¶
High risk
† Data provided by W.V. Giannobile, DDS, DMSc (e-mail communication, February 2014). ‡ Identical to the assignment of risk based on “Pattern 3” reported by Wu and colleagues.11 § SNP: Single nucleotide polymorphism. ¶ Asterisks denote either allele at this position.
173.e3 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
eTABLE 4
Power calculation for interleukin 1 (IL-1) PST genetic test.* Aa AND AA GENOTYPE RELATIVE RISKS
POWER (%)
2.0
100
1.5
100
1.4
99
1.3
91
1.25
80
1.2
63
1.1
22
* Methods and assumptions: Software used is described in Purcell S, Cherny SS, Sham PC. Genetic power calculator: design of linkage and association genetic mapping studies of complex traits. Bioinformatics 2003;19(1):149-150 and available at: http://pngu.mgh.harvard.edu/wpurcell/gpc/cc2.html. Evaluation of “main effect” based on study’s 3,860 patients (excluding smokers and those having diabetes). Assume 14.4% (n ¼ 555) “cases” (tooth extraction) and 85.6% (3,305) “controls” (no tooth extractions) as observed in the original study population. Calculations were performed using a dominant model with disease allele frequency of 0.21 with D’ disequilibrium ¼ 1.0 with the marker locus. This yields 37.4% of participants having the IL-1 PST “high-risk” composite genotype as in the actual study population. User-defined type I error rate: a = 0.05.
-
-
-
-
-
JADA 146(3) http://jada.ada.org
March 2015 173.e4
Interleukin 1 genetic tests provide no support for reduction of preventive dental care Scott R. Diehl, PhD; Fengshen Kuo, MS, PhD; Thomas C. Hart, DDS, PhD
A
study of dental benefits claims concluded that genetic polymorphisms in the interleukin 1 (IL-1) genes IL1A and IL1B (the “PST” genetic test) identified “high-risk” patients who benefitted significantly more than “low-risk” patients from receiving a second annual preventive visit by reducing tooth extractions.1 The authors of the study concluded that the data justify providing 2 annual preventive dental visits to only patients with diabetes, smokers, or those classified as “high risk” by their genetic test.1 They further claimed that the study also supported a different IL-1 test (the “PerioPredict” genetic test) that is currently being used to determine reimbursement for levels of preventive care by one major insurer. A recent review of clinical studies2 showed that independent reanalyses led to changes in findings and conclusions different from those of the original studies over one-third of the time. Unfortunately, independent reanalyses are rarely carried out on the data reported in published studies. This raises concern that the scientific community and the general public may be accepting findings that are of questionable validity. Large studies, such as the tooth extraction study mentioned above, are time-consuming and costly, thus are unlikely
ABSTRACT Background. It has been proposed that the PST and PerioPredict genetic tests that are based on polymorphisms in interleukin 1 (IL-1) genes identify a subset of patients who experience fewer tooth extractions if provided with 2 annual preventive visits. Economic analyses indicate rationing preventive care to only “high-risk” genotypes, smokers, patients with diabetes, or combinations of these risk factors would reduce the cost of dental care by $4.8 billion annually in the United States. Methods. Data presented in the study that claimed clinical utility for the PST and PerioPredict tests were obtained for reanalysis using logistic regression to assess whether the PST genetic test, smoking, diabetes, or number of preventive visits were risk factors for tooth extraction during a span of 16 years. Consistency of risk classification by the PST (version 1) and PerioPredict (version 2) genetic tests was evaluated in different ethnic groups from the 1000 Genomes database. Results. Multivariate analyses revealed association of tooth extraction with diabetes (P < .0001), smoking (P < .0001), and number of preventive visits (P ¼ .004), but no support for the PST genetic test (P ¼ .96) nor indication that the benefit of 2 preventive visits was affected by this genetic test (P ¼ .58). Classification of risk was highly inconsistent between the PST (version 1) and PerioPredict (version 2) genetic tests. Conclusions. Two annual preventive visits were supported as beneficial for all patients, and there was no evidence that the IL-1 PST genetic test has any effect on tooth extraction risk or influences the benefits of 2 annual preventive visits. Practical Implications. Neither IL-1 PST nor PerioPredict genetic tests are useful for rationing preventive dental care. Further research is needed to identify genetic biomarkers with robust clinical validity and clinical utility to effectively personalize the practice of dentistry. Key Words. Genetic screening; genetic test; biomarker; single nucleotide polymorphism; personalized medicine; tooth extraction; tooth loss; preventive care; clinical data reanalysis. JADA 2015:146(3):164-173 http://dx.doi.org/10.1016/j.adaj.2014.12.018
Dr. Diehl is a professor, Department of Oral Biology, Rutgers School of Dental Medicine, and professor of health informatics, Rutgers School of Health Related Professions, Rutgers Biomedical Health Sciences, MSB C636, Newark, NJ 07101, e-mail [email protected]. Address correspondence to Dr. Diehl. Dr. Kuo is a senior bioinformatics scientist, Department of Oral Biology, Rutgers School of Dental Medicine, Rutgers Biomedical Health Sciences, Newark, NJ. Dr. Hart is a professor, Department of Periodontics, and Director, Craniofacial Population Sciences Research, University of Illinois at Chicago College of Dentistry, Chicago, IL. He also is the chair of the ADA Council on Scientific Affairs, Chicago, IL, and will be the director of the ADA Foundation Dr. Anthony Volpe Research Center, Gaithersburg, MD, starting April 2. Copyright ª 2015 American Dental Association. All rights reserved.
164 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
to be replicated. Therefore, it is essential that complete data be made available for independent reanalysis to increase assurance of the validity of study conclusions.3 Data in the original article on risk factors for tooth extraction and patient stratification were insufficient to perform an independent reanalysis.1 Specifically, patients who have diabetes and/or were smokers—2 wellestablished risk factors for tooth loss—were pooled together within “high-risk groups” that also included patients who were classified as “high risk” based solely on their PST genotype test. Consequently, it was not possible to evaluate whether the PST genetic test itself had any effect on the clinical outcomes independent of smoking and/or diabetes. In addition to the PST test (based on 1 polymorphism in the IL1A gene and 1 in the IL1B gene), the original article also introduced a new genetic test, PerioPredict (that the authors called version 2) based on 4 other polymorphisms all located in the IL1B gene region. It was claimed that the PerioPredict test was equivalent for clinical purposes to its predecessor, the PST genetic test.1 Although the individual single nucleotide polymorphisms (SNPs) for the new PerioPredict genetic test were reported, the way by which the genotype data from the 4 SNPs were used to classify patients into “high-risk” versus “low-risk” groups was not presented in the published study.1 To perform an independent reanalysis of this study, we requested this information from the authors. We obtained data on tooth extraction outcomes separately by participants’ smoking, diabetes, PST genotype risk group, and number of preventive visits (W.V. Giannobile, DDS, DMSc, e-mail communication, December 2013). Patient age, race, or sex was not disclosed to us so these demographic measures could not be included as covariates in our reanalysis. To our knowledge, data on patients’ oral hygiene, numbers of teeth present, caries, or restorations at the start of the study period were neither obtained directly from the participants under the original study protocol nor available in the insurance database of dental procedures. In addition, the algorithm by which the 4 IL1B region SNPs were used to classify patients as “high risk” versus “low risk” for the PerioPredict test was also shared (W.V. Giannobile, DDS, DMSc, e-mail communication, February 2014). In reviewing the analyses performed in the original article based on insurance claims,1 we noted a number of potential problems and omissions of critical statistical tests needed to support the stated conclusions. As noted above, smoking and diabetes were confounded with potential effects of the IL-1 PST genetic test. Most important, although the authors claimed that “Interactions of risk status and frequency of preventive visits on tooth loss were evident .” they did not actually report results of any interaction tests.1 In fact, there was not even any direct evaluation of whether the PST genetic test had any effect at all on risk of undergoing tooth extraction (that is, whether
or not it was a “main effect”). Furthermore, no direct evaluation of whether the PST genetic test classifies individual patients as “high risk” or “low risk,” consistent with how the PerioPredict genetic test classifies the same patients, was presented. A figure in the appendix of the published study1showed comparisons of the percentage of patients with tooth loss events for “low-risk” versus “highrisk” groups and the authors of the study concluded that their version 2 (PerioPredict) test “gave results comparable with those in version 1 relative to differences between 1 and 2 preventive visits.” However, the “high-risk” group presented in their figure confounded patients who are classified as “high risk” solely because of their PST or PerioPredict genetic tests with other patients who were “high risk” because they have diabetes or were smokers. Furthermore, only group frequencies were presented in the figure and there was no way of knowing if individual patients were consistently classified as “low risk” or “high risk,” or may change their risk classification depending on which test is used. There is great potential for advances in genomics to expand our knowledge of oral disease etiology and to improve patient care through personalized approaches to diagnosis, treatment, and prevention of oral and dental diseases. However, it is essential that these powerful strategies are supported by objective and independent assessment of robust clinical data free from conflict of interest.4 Implementation of patient stratification for preventive dental care based on the IL-1 PerioPredict genetic test carries significant implications for the standard of care in dentistry. Therefore, it is essential that the data from clinical studies such as this be available for independent reanalyses. The purpose of our study was to conduct a reanalysis of data provided by the authors of the published review of dental benefits claims and discuss why our findings lead to very different conclusions. METHODS
Using only the data presented in the published study of IL1A/IL1B genotypes and tooth extractions,1 it was not possible to reanalyze the authors’ most important conclusion that an interaction exists between the PST genetic test and number of preventive visits. We obtained from W.V. Giannobile, DDS, DMSc (e-mail communication, December 2013) the numbers of participants with 1 versus 2 annual preventive visits and the ABBREVIATION KEY. ASW: African Americans from the Southwest. CEU: Caucasians from Utah. CHB: Han Chinese from Beijing. FCGR: Fc fragment of IgG, low affinity IIa or IIb, receptor. HLA: Human leukocyte antigen. IL-1: Interleukin 1. LTF: Lactotransferrin. PUR: Puerto Ricans. SNPs: Single nucleotide polymorphisms. TLR: Toll-like receptor. TNF: Tumor necrosis factor. VDR: Vitamin D receptor.
JADA 146(3) http://jada.ada.org
March 2015 165
ORIGINAL CONTRIBUTIONS
percentage in each visit group who had 1 or more tooth extractions over a period of 16 years for the following risk groups: - no risk factors; - “high-risk” IL-1 PST genetic test only; - diabetes only; - smoking only; - diabetes and smoking; - diabetes and “high-risk” IL-1 PST genetic test; - smoking and “high-risk” IL-1 PST genetic test; - all 3 risk factors (diabetes, smoking, and “high-risk” IL-1 PST genetic test) as reported in eTable 1 (available online at the end of this article). The primary outcome of the originally published study was the binary classification of each patient as those who had no tooth extraction over a follow-up period of 16 years or those who had 1 or more tooth extractions (ADA codes D7010, D7140, D7210) during the same period.1 Therefore, we applied established methods of categorical data analysis to evaluate the hypothesis that any of the risk factors listed above interact with the number of preventive visits in altering this clinical outcome. Unlike the original study in which participants were classified into a single “high-risk” group if they had diabetes, were smokers, or had been identified as having the “high-risk” IL-1 PST genotypes (thus, confounding effects of these potential risk factors), we first evaluated each risk factor separately by performing univariate logistic regression. We tested the hypotheses as to whether diabetes, smoking, number of preventive visits, or the IL-1 PST genetic test were independently associated as main effects with the tooth extraction clinical outcome. Next, we conducted multivariate logistic regression that included all 4 main effects simultaneously in the model to determine if their effects act independently. Lastly, we tested all 2-way interactions of risk factors including interaction with number of preventive visits with the IL-1 PST genetic test that was used for all of the statistical analyses reported in the dental insurance claims study. Interactions were tested in models that contained all main effects and 1 interaction term at a time in the model. Three-way interactions were not tested as sample size (especially patients with diabetes and smokers, as shown in the results) was too small to have any reasonable power. All analyses in this study were performed using JMP Genomics version 6.1 and SAS version 9.4 computer programs (SAS Institute, Cary, NC). The traditional P value of < .05 was used to assess whether an outcome was statistically significant; when we obtained much smaller P values, this was considered when interpreting the findings. We next evaluated subgroups of patients to see whether interaction between the IL-1 PST genetic test and number of preventive visits might be found within 1 of the subgroups even if not present in the sample as a whole. For example, in patients who neither had diabetes nor were smokers, we tested whether being classified as
166 JADA 146(3) http://jada.ada.org
March 2015
“high risk” or “low risk” by the IL-1 PST genetic test interacts with the benefits of a second annual preventive visit. Similar subgroup analyses were performed for patients who had diabetes, were smokers, or both had diabetes and were smokers. All tests of statistical significance reported in Giannobile and colleagues’1 study were based on what they called “version 1” of their IL-1 composite genotype test (also known as the “PST” test, abbreviated here as “PST genetic test”). They also introduced a different IL-1 genetic test that they called “version 2” (also known as the “PerioPredict” test, abbreviated here as the “PerioPredict genetic test”) based on 4 different SNPs. However, as noted above, the information presented in their publication was insufficient to classify participants as “high risk” versus “low risk” for the IL-1 PST (version 1) or the PerioPredict (version 2) genetic tests in independent samples. We requested and obtained this information from W.V. Giannobile, DDS, DMSc (e-mail communication, February 2014) as reported in eTables 2 and 3 (available online at the end of this article). Frequencies of “high-risk” versus “low-risk” genotypes were compared for the PST and PerioPredict genetic tests in 61 African Americans recruited in the Southwest United States (ASW), 55 Puerto Ricans (PUR) recruited in Puerto Rico, 55 Caucasians from Utah (CEU), and 97 Han Chinese recruited in Beijing (CHB) using data from the 1000 Genomes database Phase 1 release.5 The objectives were to measure variation in the frequency of “high-risk” participants in different populations and ethnic groups for each test and to determine how consistently the 2 different tests classified participants as high versus low risk within each group. RESULTS
We first evaluated all participants in a combined analysis to determine which risk factors showed evidence of association with the clinical outcome, tooth extraction. Odds ratios shown in Table 1 for univariate analyses revealed significant increased risk of undergoing the tooth extraction outcome with the main effects: diabetes (P < .0001), smoking (P < .0001), and decreased risk associated with 2 annual preventive visits (P ¼ .0004). In contrast, there was no evidence of association of the IL-1 PST genetic test with the tooth extraction outcome (P ¼ .8604). Multivariate analyses, in which all main effect variables are included in the model simultaneously, were performed to see if these risk factors act independently of each other. Essentially identical results were obtained, indicating that effects of diabetes, smoking, and number of preventive visits were each independently associated with the tooth extraction outcome, while the IL-1 PST genetic test still showed no evidence of significance (P ¼ .9568). Next, we tested if any risk factors gave different results depending on whether patients received 1 or 2 annual
ORIGINAL CONTRIBUTIONS
preventive visits or if other combinations of variables showed evidence of interaction. Interaction term parameters, standard errors, and P values are shown in Table 1. The results indicate no evidence that the benefits of 2 preventive visits differed depending on patients’ other risk factors. This indicates no support for interaction of number of preventive visits with diabetes (P ¼ .8074), smoking (P ¼ .4236), or the IL-1 PST genetic test (P ¼ .5788). No combination of other pairs of variables showed evidence of interaction. Results of stratified analyses were fully consistent with these findings. Three-quarters of patients in the study neither had diabetes nor were smokers. These included 1,186 participants with 1 annual preventive visit and 2,674 participants with 2 visits (Figure 1). In this subgroup, patients who received 2 annual preventive visits had lower levels of tooth extraction over a period of 16 years (P ¼ .008). The IL-1 PST genetic test had no effect on overall risk of undergoing tooth extraction (P ¼ .589) in this group of participants who neither have diabetes nor were smokers. Furthermore, the benefit of having 2 annual preventive visits was not influenced by whether a patient was classified as either “high risk” or “low risk” by the IL-1 PST genetic test (that is, there was no evidence of an interaction between visit number and the genetic test, P ¼ .475). These results of Figure 1 for patients who were nonsmokers and did not have diabetes are presented in additional detail in the first two rows of Table 2. Other subgroups shown in Table 2 include patients who had diabetes but were not smokers, smokers who did not have diabetes, and participants who both had diabetes and were smokers. There was no evidence for an effect of the PST genetic test on the tooth extraction clinical outcome in any of these subgroups (P values ¼ .814, .606, and .258, respectively) nor any support for the hypothesis that the benefit of 2 annual preventive visits in reducing tooth extraction is influenced by this genotype test (interaction P values ¼ .251, .568, and .836, respectively). It should be noted that the observed relationships between the IL-1 PST genetic test and tooth loss is highly inconsistent among patients with diabetes and smokers. For example, a higher percentage of patients who had diabetes but were nonsmokers had tooth extractions when they received 2 annual preventive visits and were “high risk” according to the PST genetic test (24.1%) versus having only 1 visit (19.0%). However, for patients who were smokers but did not have diabetes, the trend went in the opposite direction for the PST genetic test “high-risk” group (21.5% with 2 visits versus 31.0% with 1 visit). However, none of these differences approached statistical significance and were likely due to sampling error expected because of the small number of participants in these groups. Odds ratios and 95% confidence intervals (CI) for each risk factor are shown in Figure 2. No interaction term was included in this model because none was
TABLE 1
Logistic regression analysis of potential risk factors for tooth loss over 16 years. UNIVARIATE ANALYSES* P Value
Main Effects
Odds Ratio
95% Confidence Interval
IL-1† PST genetic test “high risk”
1.01
0.87-1.18
.8604
Diabetes
1.71
1.34-2.18
< .0001
Smoking
1.85
1.56-2.19
< .0001
2 visits
0.76
0.65-0.88
.0004
MULTIVARIATE ANALYSES* P Value
Main Effects
Odds Ratio
95% Confidence Interval
IL-1 PST genetic test “high risk”
1.00
0.86-1.17
Diabetes
1.66
1.28-2.10
.0001
Smoking
1.84
1.55-2.18
< .0001
.9568
0.76
0.65-0.88
.0005
Interactions‡
Estimate
Standard Error
P Value
Diabetes X visit
0.02
0.06
.8074
Smoking X visit
0.04
0.05
.4236
IL-1 PST X visit
0.02
0.04
.5788
IL-1 PST X diabetes
0.05
0.06
.4761
IL-1 PST X smoking
0.04
0.05
.3680
Diabetes X smoking
0.01
0.07
.8755
2 visits
* Univariate models for the binary outcome of none versus 1 or more tooth extractions over a period of 16 years, as in the original publication,1 were run separately for each main effect shown with only that variable in the model. Multivariate main effects’ results are based on a single model with all variables present in the model but no interactions. † IL-1: Interleukin 1. ‡ Interaction models included all 4 main effects but only 1 interaction term in each model, run separately for each of the 6 pairwise interactions shown.
statistically significant. Having 2 annual preventive visits was protective versus 1 visit, and both smoking and diabetes significantly increased risk of undergoing tooth extraction. After accounting for the effects of diabetes and smoking, patients classified as “high risk” by the PST genetic test in the published study1 did not exhibit any elevation of risk whatsoever with an odds ratio of 1.00 (95% CI 0.86-1.17). Although all tests of statistical significance reported in the published study1 were based on the PST genetic test, Interleukin Genetics has launched commercial testing for dental insurance coverage using the PerioPredict genetic test. The PerioPredict genetic test is based on 4 entirely different genetic polymorphisms than the PST genetic test but the authors claimed “either version of the IL-1 test to classify low- and high-risk patients provides the same results with regard to the relationship of the
JADA 146(3) http://jada.ada.org
March 2015 167
% OF PATIENTS WITH TOOTH EXTRACTION
ORIGINAL CONTRIBUTIONS
18%
16%
14%
12%
10% 1 Visit IL-1 PST “High-Risk” Genotypes
2 Visits IL-1 PST "Low-Risk" Genotypes
Figure 1. Percentages (standard error) of patients who had neither diabetes nor were smokers who experienced 1 or more tooth extractions over 16 years, classified as “high risk” or “low risk” by the interleukin 1 (IL-1) PST genetic test and received either 1 or 2 annual preventive visits. Having 2 visits was protective (P ¼ .008), but the IL-1 PST genotypes had no effect on risk (P ¼ .589); there is no support for interaction between number of visits and the genotype test (P ¼ .475).
number of preventive visits and tooth loss.”1 If the 2 tests actually are equivalent, they should assign the same level of risk (that is, “high risk” or “low risk”) to each patient. To evaluate test consistency, participants from 4 ethnic groups included in the 1000 Genomes database were classified as “high risk” versus “low risk” using the risk definitions of the PST and PerioPredict genetic tests (eTables 2 and 3; available online at the end of this article). As shown in Figure 3, the percentage of patients classified as “high risk” varies widely among different ethnic groups even for the same test. Furthermore, the PST and PerioPredict tests classify as “high risk” different percentages of the same population. For example, although only 11% of the ASW group were classified as “high risk” by the PST genetic test, the percentage of “high-risk” patients in this population increased more than fivefold to 61% when classified using the PerioPredict genetic test. In contrast, CEU participants decreased from 34% classified as “high risk” by the PST test to 18% “high risk” according to the PerioPredict test. Some populations, such as the PUR participants in the study, did not show any change in frequency of “high risk” participants with 25% reported as “high risk” for both the PST and PerioPredict genetic tests (Figure 3). However, classification of individual patients was inconsistent between the 2 tests. A substantial number of participants changed from “high risk” to “low risk,” or vice versa, depending on which test was used. As shown
168 JADA 146(3) http://jada.ada.org
March 2015
in Figure 4, 33% of the PUR participants had their risk classification change either from high to low risk or from low to high risk depending on which version of these IL-1 tests was used. Over one-half of ASW participants (56%), 28% of CEU, and 13% of CHB participants were assigned different risk statuses by the PST and PerioPredict genetic tests, demonstrating that the tests were not the same and were highly inconsistent. DISCUSSION
This reanalysis of a published study using data from insurance claims found no evidence for association of the IL-1 PST genetic test with risk of undergoing tooth extraction. The well-established risks associated with diabetes6 and smoking7 were confirmed, whereas having 2 annual preventive visits versus only 1 visit was associated with reducing the odds of patients having 1 or more tooth extractions over a period of 16 years by approximately 25%. Most of the comparisons in the previously reported analysis involved contrasting participants with no risk factors versus a pool of patients who had diabetes or were smokers and/or were identified as “high risk” by the IL-1 PST genetic test, confounding these variables as a single putative “highrisk” group. Our reanalysis showed that when the PST genetic test was evaluated on its own or in a multivariate model that included terms for these other risk factors (Table 1), it has no effect at all on tooth extraction risk in this data set (OR ¼ 1.0). In the data reanalysis, for most patients in the study who neither had diabetes nor smoked, being identified by the PST genetic test as “high risk” had no significant effect on risk of undergoing tooth extraction (Figure 1, P ¼ .59) whereas 2 annual preventive visits was consistently beneficial for patients (P ¼ .008), regardless of their PST genetic test status (P ¼ .47). The claim made in the published study by Giannobile and colleagues1 that potentially could have the greatest impact on the standard of care in dentistry was that only those patients who had diabetes, were smokers, or were classified as “high risk” by the IL-1 PST genetic test benefit when provided with 2 annual preventive visits by reducing risk of undergoing tooth extraction. Most
ORIGINAL CONTRIBUTIONS
TABLE 2
Stratified analyses of subgroups of patients by risk factors. RISK FACTORS
1 VISIT
Diabetes
Smoking
Interleukin 1 PST Genetic Test
No
No
Low risk
No
No
High risk
Yes
No
Low risk
No. of Participants
2 VISITS Tooth Loss (%)
DIFFERENCE (%)
Tooth Loss (%)
No. of Participants
732
16.4
1,686
13.8
2.6
454
17.0
988
12.7
4.3
67
25.4
112
18.8
6.6
Yes
No
High risk
42
19.0
79
24.1
5.0
No
Yes
Low risk
158
26.6
386
21.2
5.3
No
Yes
High risk
100
31.0
223
21.5
9.5
Yes
Yes
Low risk
19
36.8
33
27.3
9.6
Yes
Yes
High risk
12
50.0
26
34.6
15.4
P VALUES Visit
PST Genetic Test
PST Genetic Test X Visit
.008
.589
.475
.682
.814
.251
.028
.606
.568
.358
.258
.836
ODDS RATIO
patients in the study1 were classified as “low 6 risk” by the PST genetic on cti test and it was claimed tra x E 5 they would not benefit oth To g from 2 annual prevenoin erg tive visits. If this were d n 4 fU true, it would mean that ko s i R an interaction exists ed as e between the IL-1 PST r 3 Inc genetic test and number of preventive visits. 2 However, no interaction Protective No Effect tests were presented in the published study.1 In 1 our reanalysis, results of Diabetes this important interac2 Visits 1 Visit Diabetes Smoking IL-1 PST & Smoking tion test have been pre0 sented both for the RISK FACTORS FOR TOOTH EXTRACTION entire data set and in stratified analyses of subgroups. Of particular Figure 2. Odds ratios and 95% confidence intervals for each risk factor main effect in the multivariate model. Two interest are the results annual preventive visits reduced the risk of experiencing tooth loss compared with 1 annual visit. Diabetes, smoking, for the subgroup of pa- and both risk factors together significantly increased risk. The interleukin 1 (IL-1) variable showed that the “high-risk” tients who were neither IL-1 PST genetic test had no effect on risk of experiencing tooth loss in this study over a period of 16 years. smokers nor had diabetes. In no instance was there any evidence or even a 988 with 2 visits). Our finding of neither a main effect nor weak suggestion of support for this interaction. Therean interaction for this genetic test cannot be attributed to fore, our re-evaluation of this data set, using widely lack of power. In this sample, if the test had relative risk accepted and standardized methods for categorical data of 1.35 or greater, it would be expected to yield a P < .05 analysis, leads to the conclusion that a second annual over 80% of the time (eTable 4; available online at the end preventive visit was beneficial for all participants, of this article). The diabetes and smoking risk factors regardless of their IL-1 PST genetic test risk classification revealed highly significant association with tooth or whether they had diabetes or were smokers (Figure 2). extraction (Table 1) even though these groups had much There was a substantial number of patients who did smaller sample sizes (140 patients with diabetes with 1 not have diabetes and were nonsmokers with “low-risk” visit and 250 with 2 visits; 289 smokers with 1 visit, PST genotypes (732 patients with 1 visit and 1,686 with 668 with 2 visits). 2 visits) available for comparison with patients having The findings from our reanalysis are different from “high-risk” PST genotypes (454 patients with 1 visit and those of the published study. Reanalysis occurs
JADA 146(3) http://jada.ada.org
March 2015 169
ORIGINAL CONTRIBUTIONS
% OF SUBJECTS INCONSISTENT
% OF CLASSIFIED AS "HIGH RISK"
shared access to clinical study data,3,8 the focus of a project by the 70% Institute of Medicine, 61% 60% “Strategies for Responsible Sharing of Clinical 50% Trial Data.”9 Our reanalysis shows that the 40% 34% benefits of data sharing 30% extend to clinical 25% 25% studies in dentistry. 18% 20% We have also shown 11% 11% that the PST and 10% 4% PerioPredict genetic tests are different. They 0% PST (Version 1) PerioPredict (Version 2) classify risk for the INCONSISTENCY ACROSS TESTS same patient inconsistently a large portion of ASW CHB PUR CEU the time, and both tests vary substantially in the frequency of patients Figure 3. Percentages of participants classified as “high risk” by the interleukin 1 (IL-1) PST (version 1) and classified as “high risk” PerioPredict (version 2) genetic tests in African Americans from the Southwest (ASW), Puerto Ricans (PUR), across different races Caucasians from Utah (CEU), and Han Chinese from Beijing (CHB). Risk assignments differ substantially both and ethnic groups. across the different ethnic groups and between the PST and PerioPredict tests. Because of this, there are essentially no clinical data on tooth extraction frequencies 60% using the PerioPredict 56% genetic test. This raises special concern because 50% the PerioPredict genetic test is the version that is 40% now being marketed for 33% clinical practice and 28% dental insurance 30% purposes. We suggest there are 20% at least 2 reasons why 13% the IL-1 PST genetic test 10% has no effect on risk of undergoing tooth extraction. IL-1 genetic 0% ETHNIC GROUPS polymorphisms have been associated with ASW PUR CEU CHB risk of developing periodontitis at only weak to moderate strength, and periodontitis is Figure 4. Percentages of individual participants whose risks are classified differently by the interleukin 1 (IL-1) PST only 1 of many causes (version 1) and PerioPredict (version 2) genetic tests in African Americans from the Southwest (ASW), Puerto Ricans (PUR), Caucasians from Utah (CEU), and Han Chinese from Beijing (CHB). of tooth extraction in middle-aged adults. By infrequently in medical and dental research but a recent contrast, 2 underlying assumptions of the originally review of 37 reanalyses of clinical studies found that 35% published study were that the PST genetic test identifies of these led to conclusions different from the original subjects who have substantially increased risk of periodontitis, and that periodontitis, in turn, is the major report.2,3 This shows the merit in providing broad,
170 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
IL-1 PST
Tooth Extraction
Periodontitis
A
Smoking
Gender Diabetes Host Genes: IL1 IL2 IL6 IL10 FCGRs HLA LTF TLRs TNFs VDR and many other genes
Age
Oral Microbiome
Alcohol
Caries Tooth Extraction
Periodontitis
Smoking
Oral Microbiome
Failed Restorations
Root Canal Host Genes Trauma
B Figure 5. A. Simple relationships between interleukin 1 (IL-1) PST genotypes and risk of developing periodontitis and tooth extraction. B. More realistic model of the complex etiologies with multiple risk factors for periodontitis and multiple causes of tooth extraction illustrating that periodontitis is only 1 of many risk factors for this outcome. FCGR: Fc fragment of IgG, low affinity IIa or IIb, receptor. HLA: Human leukocyte antigen. LTF: Lactotransferrin. TLR: Toll-like receptor. TNF: Tumor necrosis factor. VDR: Vitamin D Receptor.
cause of tooth extraction in middle-aged adults (Figure 5A). In fact, the situation is far more complex as illustrated in Figure 5B. Association studies of the PST “high risk” genotypes (and numerous other IL-1 “composite genotypes” and individual SNPs reported in the literature over the years) with chronic periodontitis indicates that this inherited variation has at most only a weak-tomoderate effect on disease risk.10 Even in studies with highly-significant P values, many periodontitis cases do not have the “high-risk” genotype and many health controls do have this putative “high-risk” genotype. For example, among patients who did not have diabetes and were nonsmokers, 30% of chronic periodontitis cases and 20% of controls were classified as “high risk” by the IL-1 PerioPredict genetic test as reported in a recent study (Table 5 in Wu and colleagues11 where the PerioPredict genetic test is labeled “Pattern 3”). Even though the association is statistically significant (P ¼ .005), no clinically useful information is provided by the test. Furthermore, none of the genomewide association studies of periodontitis have revealed any evidence that polymorphisms in IL-1 genes are ranked anywhere among the best-supported SNPs for disease susceptibility.12-15 Thus, evidence indicates that inherited genetic variation in other genes and chromosome regions accounts for most of the large heritable component of periodontitis (Figure 5B). Furthermore, environmental
factors also play an important role in disease susceptibility.16 It is well established that tooth loss is related to age, sex, percentage of teeth with restorations, alcohol, smoking, diabetes, and number of teeth present at the start of a study.17 In addition, as we have shown here, preventive dental care also protects against tooth loss. Because periodontitis is 1 cause, but only 1 of many causes of tooth extraction (Figure 5B), it is hardly surprising that the IL-1 PST or PerioPredict genetic tests for genotype had no strong predictive association with tooth extraction as our reanalysis indicates. The data provided to us from the authors of the published study were limited to the dichotomized outcome of patients having no tooth extractions versus having 1 or more tooth extractions over a period of 16 years. Although only roughly 1 in 5 patients lost any teeth, it is possible that the small percentage of patients who lost large numbers of teeth over the 16 years might have an especially strong association with number of preventive visits or with 1 of the other risk factors considered in the study, but that possibility could not be considered in our reanalysis. SNP-based genetic testing for common complex diseases such diabetes, mental illness, cardiovascular disease, and obesity is currently of limited clinical utility, as these tests are generally no better at predicting disease risk than existing blood tests and family history.18,19 Concerns have been raised by a number of advisory
JADA 146(3) http://jada.ada.org
March 2015 171
ORIGINAL CONTRIBUTIONS
groups including the US Government Accountability Office20; the Secretary’s Advisory Committee on Genetic Testing21; the Evaluation of Genomic Applications in Practice and Prevention Working Group22; and the National Human Genome Research Institute23 regarding the limited clinical utility of SNP-based genetic testing for common, complex diseases. These bodies have noted a lack of regulation and particular concern that demonstration of clinical validity and clinical utility of such tests are not required before tests are marketed. Studies need independent replication and demonstration of clinical validity and utility as well as training of health professionals in understanding the meaning of test results so that they can explain the results clearly at a level appropriate for their patients. In Mendelian diseases such as cystic fibrosis, amelogenesis imperfecta, and some forms of tooth agenesis, mutation of a single gene causes the disease.24 In contrast, it is now widely accepted among human geneticists that heritability for common complex diseases such as type 2 diabetes, rheumatoid arthritis, and myocardial infarction, with rare exceptions, is due to the accumulated effects of genetic variation at hundreds or thousands of genes. Each variant has a weak to moderate effect, similar in magnitude to that of the weak to moderate effect of IL-1 SNPs on risk of developing chronic periodontitis.25 Because so many DNA variants are involved in risk of developing complex disease, and because there are often important environmental and epigenetic effects too, even when individual SNPs are disease-associated with strong statistical significance, they have little ability to predict disease risk in patients. For example, the SNP rs10993994 in the MSMB gene is associated with risk of developing prostate cancer (P value of 8.7 10-29).26 This gene is synthesized by the epithelial cells of the prostate gland and has effects on cell proliferation and apoptosis, making it a good candidate gene for prostate cancer. However, the “high-risk” C allele at this strongly associated SNP has a frequency of 38% in controls and 46% in cases. Knowing which of these alleles a patient has does not provide clinically actionable information either for the patient or for their doctor. Because the SNP is only 1 of many etiologic risk factors, and accounts for only a small amount of overall disease risk, people often develop disease with or without this specific SNP. Thus, using this individual SNP as a “stand-alone” test is not clinically useful for disease risk prediction. In spite of the great challenges of complex diseases, progress is being steadily made. Genomic technologies are continuing to rapidly advance, with high-throughput sequencing of DNA and RNA, proteomic, and epigenetic profiling providing increasing coverage at lower costs. Importantly, bioinformatics and complex statistical strategies and approaches, often involving high-performance computing, are also enhancing clinical
172 JADA 146(3) http://jada.ada.org
March 2015
researchers’ capabilities. Success will also clearly require assembly of much larger numbers of patients and controls than employed in most dental genetic studies conducted to date. One recent example that may suggest a path ahead is a study of type 1 diabetes that included 4,574 cases and a genetic score based on summation of SNPs at 40 non-HLA (human leukocyte antigen) genes that have been identified from replicated large genomewide studies.27 Use of the genetic score significantly improved disease prediction beyond that provided by HLA alone. That study27 and others using genetic score approaches including both modifiable and nonmodifiable risk factors28 are showing promise, but even with dozens of SNPs, these approaches are still not ready for use in the clinical setting for disease prediction. Finally, it should be noted that even when SNPs are not useful for disease prediction, genetic studies of complex diseases have already provided many important insights into previously unknown but etiologically important biological pathways that point toward novel approaches for disease prevention and therapeutic targets.29 CONCLUSIONS
Our reanalysis of a tooth extraction study1 that used insurance claims spanning 16 years has led to conclusions different from those reported in the original article. The data provide support for 2 annual preventive visits as beneficial for all patient groups and confirm the well-established effects of smoking and diabetes as risks for tooth extraction in adults. In contrast, the IL-1 PST genetic test has no significant relationship to tooth extraction risk, and there is no evidence that the benefits of having 2 annual preventive visits differ between “high-risk” and “low-risk” genotype groups. The IL-1 PST and PerioPredict genetic tests are not equivalent, as they classify patients’ risks differently, and neither test has any evidence supporting their use for patient stratification or any other aspect of dental care. n SUPPLEMENTAL DATA
Supplemental data can be found at: http://dx.doi.org/ 10.1016/j.adaj.2014.12.018. Disclosure. None of the authors reported any disclosures. The authors thank Olga Korczeniewska for assistance with initial data analyses and graphics, and Rushi Patel for help preparing tables and figures. 1. Giannobile WV, Braun TM, Caplis AK, Doucette-Stamm L, Duff GW, Kornman KS. Patient stratification for preventive care in dentistry. J Dent Res. 2013;92(8):694-701. Available at: http://jdr.sagepub.com/content/92/8/ 694.full. Accessed January 20, 2015. 2. Ebrahim S, Sohani ZN, Montoya L, et al. Reanalyses of randomized clinical trial data. JAMA. 2014;312(10):1024-1032. 3. Krumholz HM, Peterson ED. Open access to clinical trials data. JAMA. 2014;312(10):1002-1003. 4. Offit K. Personalized medicine: new genomics, old lessons. Hum Genet. 2011;130(1):3-14.
ORIGINAL CONTRIBUTIONS
5. 1000 Genomes Project Consortium; Abecasis GR, Auton A, Brooks LD, et al. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491(7422):56-65. 6. Kapp JM, Boren SA, Yun S, LeMaster J. Diabetes and tooth loss in a national sample of dentate adults reporting annual dental visits. Prev Chronic Dis. 2007;4(3):A59. 7. Hanioka T, Ojima M, Tanaka K, Matsuo K, Sato F, Tanaka H. Causal assessment of smoking and tooth loss: a systematic review of observational studies. BMC Public Health. 2011;11:221. 8. Tucker ME. How should clinical trial data be shared? BMJ. 2013;347: f4465. 9. Institute of Medicine. Sharing Clinical Trial Data: Maximizing Benefits, Minimizing Risk. Washington, DC: The National Academies Press; 2015. 10. Diehl SR, Chou CH, Kuo F, Huang C-Y, Korczeniewska O. Genetic susceptibility to periodontal disease. In: Newman MG, Takei HH, Klokkevold PR, Carranza FA, eds. Carranza’s Clinical Periodontology. 12th ed. St. Louis, MO: Elsevier Saunders; 2014:101-115. 11. Wu X, Offenbacher S, López NJ, et al. Association of interleukin-1 gene variations with moderate to severe chronic periodontitis in multiple ethnicities. J Periodontal Res. 2015;50(1):52-61. 12. Schaefer AS, Richter GM, Nothnagel M, et al. A genome-wide association study identifies GLT6D1 as a susceptibility locus for periodontitis. Hum Mol Genet. 2010;19(3):553-562. 13. Divaris K, Monda KL, North KE, et al. Exploring the genetic basis of chronic periodontitis: a genome-wide association study. Hum Mol Genet. 2013;22(11):2312-2324. 14. Teumer A, Holtfreter B, Völker U, et al. Genome-wide association study of chronic periodontitis in a German population. J Clin Periodontol. 2013;40(11):977-985. 15. Rhodin K, Divaris K, North KE, et al. Chronic periodontitis genomewide association studies: gene-centric and gene set enrichment analyses. J Dent Res. 2014;93(9):882-890. 16. Genco RJ, Borgnakke WS. Risk factors for periodontal disease. Periodontol 2000. 2013;62(1):59-94. 17. Copeland LB, Krall EA, Brown LJ, Garcia RI, Streckfus CF. Predictors of tooth loss in two US adult populations. J Public Health Dent. 2004;64(1): 31-37.
18. Ginsburg GS, Shah SH, McCarthy JJ. Taking cardiovascular genetic association studies to the next level. J Am Coll Cardiol. 2007;50(10):930-932. 19. Wray NR, Yang J, Hayes BJ, Price AL, Goddard ME, Visscher PM. Pitfalls of predicting complex traits from SNPs. Nat Rev Genet. 2013;14(7):507-515. 20. Kuntz G. Direct-to-consumer genetic tests: misleading test results are further complicated by deceptive marketing and other questionable practices. Testimony before the subcommittee on oversight and investigations, committee on energy and commerce, House of Representatives; 2010. Washington, DC: US Government Accountability Office Publication GAO-10-847T. 21. Secretary’s Advisory Committee on Genetic Testing, National Institutes of Health. Enhancing the oversight of genetic tests: recommendations of the SACGT; 2000. Available at: http://osp.od.nih.gov/sites/default/files/ resources/oversight_report.pdf. Accessed February 3, 2015. 22. Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group. Recommendations from the EGAPP Working Group: does genomic profiling to assess type 2 diabetes improve health outcomes? Genet Med. 2013;15(8):612-617. 23. National Human Genome Research Institute, National Institutes of Health. Regulation of genetic tests. Available at: www.genome.gov/1 0002335. Updated September 2, 2014. Accessed November 6, 2014. 24. US National Library of Medicine, National Institutes of Health. Genetics Home Reference; 2014. Available at: http://ghr.nlm.nih.gov/. Accessed November 6, 2014. 25. Ripke S, O’Dushlaine C, Chambert K, et al. Genome-wide association analysis identifies 13 new risk loci for schizophrenia. Nat Genet. 2013;45(10): 1150-1159. 26. Liu H, Wang B, Han C. Meta-analysis of genome-wide and replication association studies on prostate cancer. Prostate. 2011;71(2):209-224. 27. Winkler C, Krumsiek J, Buettner F, et al. Feature ranking of type 1 diabetes susceptibility genes improves prediction of type 1 diabetes. Diabetologia. 2014;57(12):2521-2529. 28. Langenberg C, Sharp SJ, Franks PW, et al. Gene-lifestyle interaction and type 2 diabetes: the EPIC interact case-cohort study. PLoS Med. 2014; 11(5):e1001647. 29. Diogo D, Okada Y, Plenge RM. Genome-wide association studies to advance our understanding of critical cell types and pathways in rheumatoid arthritis: recent findings and challenges. Curr Opin Rheumatol. 2014;26(1):85-92.
JADA 146(3) http://jada.ada.org
March 2015 173
ORIGINAL CONTRIBUTIONS
eTABLE 1
Participant counts and frequency of tooth extraction (loss) by risk factor group and number of preventive visits.* NO. OF RISK FACTORS
RISK FACTOR
1 VISIT No. of Participants
2 VISITS Tooth Loss (%)
No. of Participants
TOTAL Tooth Loss (%)
0
None
732
16.4
1,686
13.8
2,418
1
Diabetic only
67
25.4
112
18.8
179
1
Smoking only
158
26.6
386
21.2
544
1
IL-1† PST
454
17.0
988
12.7
1,442
2
Diabetic and smoking
19
36.8
33
27.3
52
2
Diabetic and IL-1 PST
42
19.0
79
24.1
121
2
Smoking and IL-1 PST
100
31.0
223
21.5
323
3
Diabetic, smoking, and IL-1 PST
12
50.0
26
34.6
39
TOTAL
All Risk Factor Groups
1,584
19.4
3,533
15.4
5,119
* Data provided by W.V. Giannobile, DDS, DMSc (e-mail communication, December 2013). † IL-1: Interleukin 1.
173.e1 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
eTABLE 2
Classification of “high risk” and “low risk” by interleukin 1 PST (version 1) genetic test† GENE
IL1A
IL1B
SNP LOCUS‡
4845
3954
ALLELES
G/T
C/T
rs NUMBER
17561
1143634
Genotypes
T/*
§
TEST RESULT
T/*
High risk
T/*
C/C
Low risk
G/G
T/*
Low risk
G/G
C/C
Low risk
† Data provided by W.V. Giannobile, DDS, DMSc (e-mail communication, February 2014). ‡ SNP: Single nucleotide polymorphism. § Asterisks denote either allele at this position.
JADA 146(3) http://jada.ada.org
March 2015 173.e2
ORIGINAL CONTRIBUTIONS
eTABLE 3
Classification of “high risk” and “low risk” by interleukin 1 PerioPredict (version 2) genetic test.†‡ GENE
IL1B
IL1B
IL1B
IL1B
SNP LOCUS§
-511
-1464
-3737
3877
ALLELES
TEST RESULT
C/T
G/C
C/T
G/A
rs NUMBER
16944
1143623
4848306
1143633
Genotypes
C/C
G/G
C/C
*/*
C/C
G/G
T/T
G/G
High risk
C/T
G/C
C/C
*/*
High risk
C/T
G/G
C/C
*/*
High risk
C/T
G/G
C/T
G/G
High risk
T/T
G/*
C/C
*/*
High risk
C/C
C/*
*/*
*/*
Low risk
C/C
G/G
C/T
*/*
Low risk
C/C
G/G
T/T
A/*
Low risk
C/T
C/C
*/*
*/*
Low risk
C/T
G/C
T/*
*/*
Low risk
C/T
G/G
C/T
A/*
Low risk
C/T
G/G
T/T
*/*
Low risk
T/T
C/C
*/*
*/*
Low risk
T/T
G/*
T/*
*/*
Low risk
¶
High risk
† Data provided by W.V. Giannobile, DDS, DMSc (e-mail communication, February 2014). ‡ Identical to the assignment of risk based on “Pattern 3” reported by Wu and colleagues.11 § SNP: Single nucleotide polymorphism. ¶ Asterisks denote either allele at this position.
173.e3 JADA 146(3) http://jada.ada.org
March 2015
ORIGINAL CONTRIBUTIONS
eTABLE 4
Power calculation for interleukin 1 (IL-1) PST genetic test.* Aa AND AA GENOTYPE RELATIVE RISKS
POWER (%)
2.0
100
1.5
100
1.4
99
1.3
91
1.25
80
1.2
63
1.1
22
* Methods and assumptions: Software used is described in Purcell S, Cherny SS, Sham PC. Genetic power calculator: design of linkage and association genetic mapping studies of complex traits. Bioinformatics 2003;19(1):149-150 and available at: http://pngu.mgh.harvard.edu/wpurcell/gpc/cc2.html. Evaluation of “main effect” based on study’s 3,860 patients (excluding smokers and those having diabetes). Assume 14.4% (n ¼ 555) “cases” (tooth extraction) and 85.6% (3,305) “controls” (no tooth extractions) as observed in the original study population. Calculations were performed using a dominant model with disease allele frequency of 0.21 with D’ disequilibrium ¼ 1.0 with the marker locus. This yields 37.4% of participants having the IL-1 PST “high-risk” composite genotype as in the actual study population. User-defined type I error rate: a = 0.05.
-
-
-
-
-
JADA 146(3) http://jada.ada.org
March 2015 173.e4