- Email: [email protected]
Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS Vaccine xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling Ran D. Balicer a,b,∗ , Chandra J. Cohen a , Morton Leibowitz a , Becca S. Feldman a , Ilan Brufman a , Craig Roberts c , Moshe Hoshen a a
Clalit Research Institute, Chief Physician’s Office, Clalit Health Services, Arlozorov 101, Tel Aviv, Israel Department of Epidemiology, Faculty of Health Sciences, Ben Gurion University, Be’er Sheva, Israel c Pfizer Inc., Outcomes Research, 500 Arcola Road, Collegeville, PA 19301, USA b
a r t i c l e
i n f o
Article history: Received 11 March 2013 Received in revised form 12 November 2013 Accepted 10 December 2013 Available online xxx Keywords: Pneumococcal disease Pneumonia Vaccine strategy Risk profiling
a b s t r a c t Background: Current pneumococcal vaccine campaigns take a broad, primarily age-based approach to immunization targeting, overlooking many clinical and administrative considerations necessary in disease prevention and resource planning for specific patient populations. We aim to demonstrate the utility of a population-specific predictive model for hospital-treated pneumonia to direct effective vaccine targeting. Methods: Data was extracted for 1,053,435 members of an Israeli HMO, age 50 and older, during the study period 2008–2010. We developed and validated a logistic regression model to predict hospitaltreated pneumonia using training and test samples, including a set of standard and population-specific risk factors. The model’s predictive value was tested for prospectively identifying cases of pneumonia and invasive pneumococcal disease (IPD), and was compared to the existing international paradigm for patient immunization targeting. Results: In a multivariate regression, age, co-morbidity burden and previous pneumonia events were most strongly positively associated with hospital-treated pneumonia. The model predicting hospital-treated pneumonia yielded a c-statistic of 0.80. Utilizing the predictive model, the top 17% highest-risk within the study validation population were targeted to detect 54% of those members who were subsequently treated for hospitalized pneumonia in the follow up period. The high-risk population identified through this model included 46% of the follow-up year’s IPD cases, and 27% of community-treated pneumonia cases. These outcomes were compared with international guidelines for risk for pneumococcal diseases that accurately identified only 35% of hospitalized pneumonia, 41% of IPD cases and 21% of communitytreated pneumonia. Conclusions: We demonstrate that a customized model for vaccine targeting performs better than international guidelines, and therefore, risk modeling may allow for more precise vaccine targeting and resource allocation than current national and international guidelines. Health care managers and policymakers may consider the strategic potential of utilizing clinical and administrative databases for creating population-specific risk prediction models to inform vaccination campaigns. © 2013 Elsevier Ltd. All rights reserved.
1. Introduction Pneumococcal disease is responsible for a heavy burden of disease and mortality globally [1,2]. Invasive pneumococcal disease (IPD) is a critical and often fatal expression of pneumococcal
∗ Corresponding author at: Clalit Research Institute, Chief Physician’s Office, Clalit Health Services, Arlozorov 101, Tel Aviv, Israel. Tel.: +972 36923104; fax: +972 36925821. E-mail addresses: [email protected] (R.D. Balicer), [email protected] (C.J. Cohen), [email protected] (B.S. Feldman), [email protected] (I. Brufman), Craig.Roberts@pfizer.com (C. Roberts), [email protected] (M. Hoshen).
diseases, albeit relatively rare [3,4]. Pneumococcal pneumonias are more common and represent substantial costs from hospitalization and morbidity [5–8]. It is, however, difficult to identify the causative agent of pneumonia and to pinpoint those cases triggered by Streptococcus pneumonia [8]. Consequently, preventive strategies generally take a broad stroke approach to targeting risk for pneumonia, even though international guidelines issued by the Centers for Disease Control and Prevention’s Advisory Committee on Immunization Practices (ACIP) for identifying high-risk individuals have been based on research that identified risk of IPD [4,10]. While vaccines are considered a first line of defense against pneumococcal disease, the widely-used pneumococcal polysaccharide vaccine (PPSV23) is currently under scrutiny for its clinical
0264-410X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.vaccine.2013.12.020
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
2
effectiveness [1,11–13]. Much of the evidence in the literature supports PPSV23’s protective efficacy against IPD [1,3,10,12]; yet, evidence on its effectiveness in preventing pneumonias and other pneumococcal infections yields inconclusive results [4]. In parallel, a conjugate pneumococcal vaccine, considerably more costly, is an emerging alternative for pneumococcal disease prevention in adults. With the clinical efficacy of pneumococcal vaccines in question for preventing all-cause pneumonia and the FDA’s recent approval of a conjugate vaccine for use in older adults, aged 55 and up [14], policy-makers and health care managers are facing decisions about the extent of coverage and appropriate target populations for pneumococcal vaccines. Current national and international guidelines for administering pneumococcal vaccines are limited, usually based on broad age, chronic disease and immune-suppression criteria, yielding large, non-specific target groups [9,15]. While these universal clinical groupings are a valid starting point for vaccine targeting strategies, a more discriminative identification system is desirable to allow for effective resource allocation and to reach the most appropriate sub-groups at risk. The widespread integration of electronic medical records and large health information technology (IT) databases is being harnessed to support vaccine decision making and planning efforts [16]. Health IT has been incorporated into vaccine planning for aims such as: the identification of vaccine-attributable risk [17], immunization surveillance [18], and monitoring vaccine effectiveness [19]. Clalit Health Services (Clalit), the largest health fund in Israel, has the capacity to employ its electronic medical records database with integrated administrative and clinical (hospital and community-based) medical records to achieve more targeted vaccine planning. The health fund currently uses broadly-defined risk groups based on age and ACIP recommendations to target pneumococcal vaccine administration, similar to many health organizations worldwide. We propose that building a customized predictive model to identify sub-populations at risk for acquiring pneumococcal disease is an innovative approach to augmenting vaccine management. We present in this paper the development and validation of a population-specific predictive model for hospitaltreated pneumonia utilizing Clalit’s integrated longitudinal medical records data, and test the model in predicting invasive pneumococcal disease and community-treated pneumonia. 2. Materials and methods 2.1. Setting In Israel, membership in one of four nationwide health funds is universal and mandatory. Clalit is the largest of these funds, covering 53% of Israel’s population (4 million members) and serves as both an insurer and a health care provider supplying and financing services to members within its system. We used Clalit’s comprehensive database which is founded on both administrative claims data and clinical electronic health records, including data from laboratories, community clinics, hospitals, imaging centers, and pharmacies. Data from all medical encounters, including those in Clalit-owned hospitals and clinics, as well as those outside the Clalit system, are captured in the electronic medical records and collated in the organization’s centralized database. Laboratory and medications data are downloaded directly from the electronic database. 2.2. Study population and data collection This was a cohort study during the period January 1, 2008–December 31, 2010 using historical data collected from the Clalit Health Services’ centralized electronic database on all Clalit members age 50 and older (as of January 1, 2008). Those
members who switched funds (<1%) were excluded. The study period was divided into a baseline period (January 1, 2008–December 31, 2008), in which patient characteristics were recorded, and a follow up period from January 1, 2009–December 31, 2010 in which outcomes were documented. 2.3. Variables Potential risk factors from January 1, 2008 were incorporated into the model including: area-level socio-economic status (SES), ethnicity (defined by the patient’s designated primary care clinic), age, and gender. Clinical input variables comprised hospital-treated pneumonia events and community-treated pneumonia events in 2008, and smoking status (earliest available during the study period, with indicators more extensively available from 2009 onwards). To account for immunosuppression and susceptibility to pneumonia, disease-specific risk categories for pneumococcal immunization developed by the ACIP [3] and adapted by Clalit were included. This aggregate guideline-based categorization classifies patients into one of three groups: (1) low risk: Immunocompetent subjects without any major chronic medical conditions; (2) moderate risk: Immunocompetent with one or more chronic conditions of the heart, lung, liver and metabolism; (3) high risk: Immunocompromised patients including those with hematological conditions, functional or anatomic asplenia (patients recommended by ACIP to receive 2 vaccine doses) [20]. This classification system is based on the identification of chronic conditions that are either marked by impaired immune-competence or are known to predispose to pneumococcal disease. A complete list of these chronic diseases can be found in Supplement 1. We also integrated the Johns Hopkins Adjusted Clinical Groups (ACG) co-morbidity score into the model to measure co-morbidity burden at the individual patient level [21]. The ACG system is a well-tested and validated case-mix instrument for measuring patient complexity through the classification of co-morbidities based on assumed use of resources and costs. It categorizes morbidity by grouping individuals according to their age, gender and all known medical diagnoses (assigned over a defined period of time, typically one year) [22]. To facilitate modeling, the ACG System automatically assigns a sixlevel morbidity categorization termed Resource Utilization Bands (RUBs) [21]. 2.4. Outcomes Outcomes were episodes of hospital-treated all-cause pneumonia (HTP), community-treated all-cause pneumonia (CTP) and episodes of invasive pneumococcal disease. These events were identified through clinical and administrative records, from hospital diagnoses (ICD-9 coding), laboratory records (IPD cases based on positive microbiological blood tests taken from hospital laboratory records), and community health records (ICD-9 coding). Outpatient diagnoses made within 30 days of hospitalization or of previous outpatient diagnosis were counted as part of the previous hospitalized/non-hospitalized event, respectively. If more than 30 days after hospitalization the outpatient diagnosis was counted as a separate event. If an event included inpatient and outpatient diagnoses of pneumonia, it was defined as HTP. 2.5. Analysis Univariate analyses were performed to determine the independent risk factors that are associated with HTP incidence in the outcome period (2009–2010). Multivariate logistic regression was conducted to develop an adjusted model for HTP. The population age 50+ was randomly split in half for training and validating the logistic regression model. This model using 2008 baseline data
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
3
Table 1 Characteristics (as of 1 January 2008, unless otherwise stated) and relative risk for hospital-treated pneumonia (HTP) of total Clalit study population and of patients diagnosed with HTP during 2009–2010.
N Age Age group
Sex Ethnic group
Low SES indicatora New immigrant status Smoking status (as of end of 2009)
ACG morbidity burden – RUB (2008)
Pneumococcal disease risk group (ACIP-based guidelines, 2008) Previous pneumonia event (in 2008)
Mean 50–64 65–74 75–84 85+ Female Male General+ Orthodox Jewish Arab No Yes No Yes Never Past smoker Current smoker Unknown 0 1 2 3 4 5 Baseline Moderateb Highc No HTP HTP No CTP CTP
Entire Clalit population (50+ at onset of study 1.1.2008)
Members that had an HTP event in 2009–2010 (among members 50+ on 1.1.2008)
Relative risk (RR)
1,053,435 65.5 (SD = 11.2) 559,919 (53.2%) 242,634 (23.0%) 185,379 (17.6%) 65,503 (6.2%) 578,021 (54.9%) 475,414 (45.1%) 943,829 (89.6%)
20,971 75.83 (SD = 11.19) 3825 (18.2%) 4484 (21.4%) 7819 (37.3%) 4843 (23.1%) 10,087 (48.1%) 10,884 (51.9%) 18,728 (89.3%)
– – Reference 2.71 6.17 10.82 Reference 1.31 Reference
109,606 (10.4%) 772,161 (73.3%) 281,274 (26.7%) 329,632 (31.3%) 723,803 (68.7%) 711,616 (67.6%) 138,653 (13.2%) 160,360 (15.2%) 42,806 (4.0%) 28,459 (2.7%) 89,983 (8.5%) 134,878 (12.8%) 586,177 (55.6%) 153,972 (14.6%) 59,966 (5.7%) 515,884 (49.0%) 361,020 (34.3%) 176,531 (16.8%) 1,043,780 (99.1%) 9,655 (0.9%) 1,048,401 (99.5%) 5,034 (0.5%)
2243 (10.7%) 12,107 (57.7%) 8864 (42.3%) 2912 (13.9%) 18,059 (86.1%) 13,653 (65.1%) 3443 (16.4%) 2637 (12.6%) 1238 (5.9%) 157 (0.7%) 1360 (6.5%) 1236 (5.9%) 8945 (42.7%) 5272 (25.1%) 4001 (19.1%) 3596 (17.1%) 9977 (47.6%) 7398 (35.3%) 19,674 (93.8%) 1297 (6.2%) 20,704 (98.5%) 267 (1.3%)
1.03 Reference 2.01 Reference 2.82 Reference 1.29 0.86 1.51 Reference 2.74 1.66 2.77 6.21 12.09 Reference 3.96 6.01 Reference 7.13 Reference 2.69
95% CI
– – 2.59–2.82 5.94–6.42 10.38–11.28 1.28–1.35
0.99–1.08 1.96–2.07 2.72–2.94 1.25–1.34 0.82–0.89 1.42–1.60 2.32–3.23 1.41–1.96 2.36–3.24 5.30–7.27 10.32–14.18 3.82–4.12 5.78–6.25 6.76–7.51 2.39–3.02
a
Exempt from the national social security contribution due to low income. b Moderate risk group includes the following chronic conditions: cardiomyopathy, heart failure, ischemic heart disease, IHSS, vavular heart disease, carotid artery disease and CVA, bronchiectasis, chronic bronchitis, COPD, cystic fibrosis and history of tuberculosis. c High risk group includes the following chronic conditions or clinical indicators: transplant, malignancy, hodgkins, lymphoma, splenic dysfunction/splenectomy, chronic renal failure, dialysis.
to predict HTP was then tested for its sensitivity in predicting incidence of HTP, IPD and CTP during 2009–2010 in a validation population. The c-statistic and sensitivity of the model were used to measure model performance for the three outcomes. The predictive ability of the Clalit model was then compared to that of the high and moderate ACIP-based risk groups to identify HTP, IPD, and CTP. To compare the accuracy of the Clalit model in identifying those members who would acquire pneumonia or IPD during the outcome period to the ACIP-based risk category criteria, we used a point in the Clalit model distribution that corresponds to the equivalent population size of the ACIP-based risk groups, which are referred to in Table 3 as vaccine strategies: ACIP-based moderate risk group comprised 51% of the population (strategy 2), and therefore, the corresponding cut point in the Clalit risk model was 51% (strategy 3). Whereas, the ACIP-based high risk group comprised only 16.8% (∼17%) of the total study population (strategy 4), therefore an equivalent proportion of the highest risk was taken at the cut point of 17% in the Clalit model (strategy 5). In evaluating different vaccine targeting strategies, we also calculated what percentage of the population would have to be targeted to yield the same 35% sensitivity as using the ACIP-based criteria (strategy 6). A more extreme cut point of 5% was also tested (strategy 7), to represent a strategy reflecting very limited available resources. The predictive measures were tested for statistical significance using the Chi-square test. SPSS V 18.0 was used for analysis. The institutional ethics committee at Clalit approved this study.
3. Results This population-specific study of risk for pneumonia and IPD is based on an analysis of data garnered from the records of 1,053,435 patients over the age of 50 as of January 1, 2008. The study population characteristics and relative risk (RR) assessments are presented in Table 1. The study population was 54.9% female, with the following age group distribution: 53.2% were 50–64 years old, 23.0% were 65–74 years old, 17.6% were 75–84 years old, and 6.2% were 85 and older. More than a quarter of the population (26.7%) was low income status, 28.4% were current or past smokers, and 10.4% were patients residing in predominantly Arab localities. Of the adult population age 50 and older, 51% met one or more of the criteria for high or moderate ACIP-based risk groups, while 36% of those age 50 through 64 met one or more of the ACIP criteria. Supplement 1 details the prevalence of each chronic condition in the study population. There were 20,917 members (2.0% of the Clalit population studied) that were identified as having had at least one hospital-treated pneumonia event in 2009–2010 (Table 1). Those who had a hospitalized pneumonia event were on average 75.8 ± 11.2 years old, 29% were past or current smokers, 10.7% were patients from Arab localities, and 42.3% were patients residing in areas of low SES. Table 1 shows that risk for hospitalized pneumonia among Clalit members increased with age: members age 65–74 had a relative risk (RR) of 2.71 (95% CI = [2.59–2.82]) compared to members 50–64; those ages 75–84 had an RR of 6.17 (95%
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
4
Table 2 Multivariate regression analysis of hospitalized pneumonia risk from 2009-2010 among patients 50 years and older in Clalit (training population). Mulitvariate Analysis n = 526,718
Sig.
Days hospitalized (2008)
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.007 0.844 0.249 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Age group Sex Ethnic group Low SES indicator New immigrant status Smoking status (2010)
ACG morbidity burden – RUB (2008)
Pneumococcal disease risk group (ACIP-based MoH definitions, 2008) Previous pneumonia event (2008)
50–64 65–74 75–84 85+ Male Arab Yes Yes Never Past smoker Current smoker Baseline 1 2 3 4 5 Baseline Moderate High Hospital-treated event Community-treated event
B
Exp(B) [OR]
95% C.I. for OR Lower
Upper
0.012
1.012
1.010
1.014
0.598 1.308 1.872 0.247 0.256 0.364 0.241
1.818 3.697 6.501 1.280 1.292 1.439 1.273
1.704 3.472 6.055 1.226 1.211 1.380 1.197
1.940 3.936 6.980 1.336 1.379 1.502 1.353
0.141 0.318
1.151 1.374
1.087 1.290
1.219 1.464
0.381 −0.028 0.156 0.485 0.782
1.464 0.973 1.169 1.625 2.187
1.110 0.738 0.896 1.243 1.670
1.929 1.281 1.523 2.123 2.863
0.761 0.915 0.943 0.583
2.141 2.496 2.568 1.792
2.015 2.335 2.334 1.488
2.275 2.669 2.827 2.159
Table 3 Strategies for vaccine targeting (test population). Vaccination strategy 1 2 3 4 5 6 7 a b c
High and Moderate risk groups and all aged 65+ High and Moderate ACIP-based risk groups Clalit model, 51% highest risk scores (based on training model) ACIP-based highest risk group (Immunosuppressed) Clalit model, 17% highest risk scores (based on training model) Clalit model, 8.6% highest risk scores (based on training model) Clalit model, 5% highest risk scores (based on training model)
% of HTP cases in 2009–2010 identified (n = 10,423)
% of IPD cases in 2009-10 identified (n = 90)
66% (347,008)
94% (9818)
89% (80)
78% (3572)
51% (268,616)
a
83% (8609)
80% (72)
65% (2980)b
51% (267,744)
85% (8896)a
80% (72)
66% (3045)b
17% (88,142)
35% (3634)c
41% (37)
21% (971)c
17% (87,853)
54% (5667)c
46% (41)
27% (1246)c
8.6% (45,521)
35% (3634)
31% (28)
15% (692)
5% (25,580)
23% (2390)
18% (16)
9% (421)
% of 50+ population targeted (n = 526,717)
% of CTP cases in 2009–2010 identified (n = 4,603)
When comparing strategies 2 and 3, p < 0.001 using McNemar test. When comparing strategies 2 and 3, p < 0.05 using McNemar test. When comparing strategies 4 and 5, p < 0.001 using McNemar test.
CI = [5.94–6.42]), and those age 85 and older had an RR of 10.82 (95% CI = [10.38–11.28]). Male gender RR = 1.31; [CI = 1.28–1.35], and morbidity burden, as measured by the ACG RUBs, had RRs reaching 12.09 [CI = 10.32–14.18] for members in the highest morbidity burden group. The risk of HTP was higher among members with low socio-economic status (SES), with an RR = 2.01 (95% CI = 1.96–2.07). In Table 2 we present the results of the multivariate logistic regression model. HTP is positively associated with age, those age 75–84 had an OR = 3.70 and those age 85+ had an OR = 6.50 (95% CI = 6.06–6.98) compared to those age 50–64. A recent history of hospital-treated or community-treated pneumonia increased the risk of another event with ORs of 2.57 (95% CI = 2.33–2.82) and 1.79 (95% CI = 2.33–2.82), respectively. ACIP-based risk groups were associated with significantly increased risk beyond many of the above risk factors (OR = 2.50 [95% CI = 2.34–2.67]) for the high risk group and OR = 2.14 [95% CI = 2.02–2.28] for the moderate risk group), and a high morbidity burden RUB status had an additional significant effect (OR = 1.63 [95% CI = 1.24–2.12] and 2.19
[95% CI = 1.67–2.86] for the highest 2 categories). While current smoking appeared to be protective for HTP in the univariate model, (RR of 0.86; 95% CI = 0.82–0.89), in the multivariate model, adjusted for multiple confounders, it becomes a significant risk factor for HTP (RR = 1.37; 95% CI = 1.29–1.47) and past smoker is a weaker yet significant risk factor (RR = 1.15; 95% CI = 1.09–1.22) compared to members who never smoked. The performance indicators for the Clalit tailored model were high for both the training and test populations, with a c-statistic of 0.80 for the test population model. In Table 3 we illustrate several vaccine targeting strategy options and the application of the HTP calibrated model to predict other related clinical syndromes, IPD and CTP. With a narrow target population of 17% (the percentage that corresponds to the ACIP-based high-risk category within the study population), sensitivity comparisons show that the ACIP-based strategy 4 yielded 35%, 41% and 21% sensitivity predicting HTP, IPD and CTP, respectively; while the Clalit tailored model strategy 5 yielded 54%, 46% and 27%, for each of the three
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
outcomes, respectively. Comparison of sensitivities measured were statistically significant for both HTP and CTP (p < 0.001) and was not statistically significant for IPD, given low incidence (p = 0.222). Looking at a broader target population, the ACIP-based strategy 2, these high/moderate risk patients comprised 51% of the entire population, and 83% of the HTP cases occurred in this population segment. Selecting 51% of the population, strategy 3, using our customized model yielded a slightly higher sensitivity of 85%, with differences between these two strategies statistically significant at p < 0.001 for HTP. In an additional analysis, predicting and identifying 35% of the HTP cases with our model would involve targeting vaccination for only 8.6% of adults age 50 and older, as compared to 17% when using ACIP-based criteria.
4. Discussion Immunization programs in many countries in Europe and North America have endorsed universal age-based pneumococcal vaccination policies among older adults (over the age of 65), as well as for immune-compromised younger adults [9,15,23,24]. In the United States, these recommendations by the ACIP are longstanding for over a decade [25]. This widespread strategy is, however, being brought into question even in countries where the vaccines were routinely administered [11]. When applied in practice, many patients, whose risk for pneumonia or IPD is modest at most, get vaccinated based solely on age. As more costly vaccines become widely available, health provider organizations in Israel and in other countries may seek to further target populations for vaccination campaigns. Our findings suggest that country-level guidelines can be tailored for an organization’s specific population characteristics, epidemiologic circumstances, and financial structures to optimize allocation of limited resources of a subsidized vaccination outreach program. In this study we document an initial effort to utilize an integrated electronic medical database for population-specific targeting in pneumococcal vaccine management to optimize disease prevention and resource planning. Risk profiles for pneumococcal disease have predominantly been supported by research focused on IPD, as it is the one illness event in which pneumococcus can definitively be identified as the causative organism. It has also been shown, however, that a large proportion of hospital treated pneumonia is caused by pneumococcus (Streptococcus pneumonia) [9,26]. Therefore, we examined risk factors for HTP in our model because the costs and burden of hospitalized pneumonias are a significant concern for pneumococcal disease prevention strategies [2,5,6]. To build a predictive model customized to Clalit’s population, we combined widely-accepted risk parameters such as age, sex, smoking status and socio-economic status (SES) with a national ACIP-based risk classification, and patient morbidity burden as measured by John Hopkins Adjusted Clinical Groups (ACG) based RUB index. This customized model elucidates the following as the most prominent risk factors for HTP among Clalit members: older age (75+), high morbidity (from both the ACG morbidity and ACIP-based risk classifications), and having experienced a previous pneumonia event. While age and previous pneumonia events have been identified in the literature as important risk factors, the use of the ACG morbidity burden indicator provides an important additional prognostic indicator beyond immunosuppression to strengthen our predictive model. Accurately predicting those individuals who will experience a future pneumonia event to enable preventative action underscores the motivation for building this model. Among Clalit’s total population 50 and older, 20,971 (2%) individuals were identified as having an HTP event, 192 having an IPD event, and 9250 having a CTP event in 2009–2010. When compared in an equivalent proportion
5
of the population (17%), the Clalit custom model exhibited higher predictive utility for HTP than using ACIP-based risk profiles, as shown in Table 3 (comparison of strategies 4 and 5). Additionally, the Clalit model would allow more judicious use of vaccines to capture the same 35% of HTP cases, targeting half as many of its 50+ adult population (8.6% vs. 17%) than if it were using ACIP-based risk criteria (comparing strategies 4 and 6). When comparing strategies toward a broader target population (strategies 2 and 3), the differences in sensitivities are statistically significant but of minimal population impact. A customized model is thus clinically relevant only when resources are considerably limited, and do not allow for vaccination of the entire high/moderate risk group determined by the ACIP criteria. The added value of using the customized model becomes apparent when greater targeting is required. Through our evaluation we have demonstrated the ability of an organization to use its database to develop a risk model with better predictive accuracy than internationally-defined criteria to effectively identify high-risk sub-groups of patients for vaccine administration targeting. 4.1. Limitations Our study has some important limitations to take into consideration. While we included a large set of risk factor variables in our logistic regression analysis, we did not incorporate an exhaustive list of factors. Previous studies on risk factor identification for pneumococcal disease have identified additional key variables to those that were part of our study, including: medications, frailty indices, allergies, pets, environmental irritants, and poor oral hygiene [27]. The predictive accuracy of our model as measured by the c-statistic 0.80 is considered robust; however with the incorporation of additional risk variables, the model can potentially provide greater predictive value. Using ecological clinical variables such as SES and ethnicity may have caused some confounding bias, thereby reducing the sensitivity of our model. 4.2. Implications for practice Utilizing this customized risk model developed within Clalit provides a predictive advantage over international guideline risk criteria, for the purpose of targeting pneumococcal vaccines. As illustrated in Table 3, specific targets for vaccination can be tailored to better balance clinical risks associated with vaccine administration. In Clalit’s population, more than 1 million members (25% of all members) are 50 years and older and almost 500.000 members are age 65 and older. In our test population (n = 5,26,717), when considering both age-based (65+) and clinical risk groupings currently recommended by the ACIP, 66% (3,47,008) of Clalit members aged 50+ meet the vaccination criteria. Ignoring the agebased recommendations and employing the ACIP clinical high and moderate-risk criteria, still over 50% of members (268,616) aged 50+ meet one or more of these criteria. Targeting only the highest ACIP-based risk group for pneumococcal disease (comprised of immune-suppressed members) for administration would entail providing vaccines for 88,142 Clalit members (17%) – potentially still a large target for a costly vaccination campaign. If this target population size was not manageable for the organizational resources available, the model could be adjusted to use a lower cut point, such as 5% (strategy 7 in Table 3). As a function of the power of a customized strategy, the levels of disease prevention remain high despite a substantial reduction in the target population size. Using clinical and administrative data to direct vaccination policy helps to prioritize organizational planning, resource allocation and clinical decision-making through the identification of key highrisk sub-groups and the monitoring of vaccination administration.
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
6
With electronic patient health records (EHRs), an organization can identify the 50+ cohort and determine risk profiles for organizationspecific vaccination guidelines. A diverse set of risk factors can be tested and added to the model, depending on availability of data. These risk factors can be integrated into an automated algorithm within the EHR database to flag patients at high risk, for practitioners accessing patients’ files or to create lists of candidates for vaccination outreach. Using our methodology, a risk cut-off threshold can be adjusted to accommodate the budget and operational needs of the organization or physician practice. 5. Conclusion A customized risk model has the potential to enhance cost management and facilitate clinical efforts by focusing costs on the individuals in the population who stand to benefit most, and targeting future vaccines to patients at highest risk. We expect that such predictive modeling will become widespread in many other clinical applications as well. We recommend that health care managers and policy-makers with access to clinical and administrative databases explore the potential for creating population-specific risk prediction models for more effective vaccination targeting. Funding This study was funded in part by an unconditional institutional research grant from Pfizer Inc. The authors did not receive any individual financial contribution from this funding source. The sponsor provided methodological comments for the study design, and critically reviewed outcome reports and the manuscript. Conflicts of interest C.R. is an employee of Pfizer Inc., and has served as a methodological adviser to the research group. None of the other authors have any conflicts of interest to declare. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.vaccine. 2013.12.020. References [1] Moberley S, Holden J, Tatham DP, Andrews RM. Vaccines for preventing pneumococcal infection in adults. The Cochrane Database of Systematic Reviews 2008;4:57 (review). [2] American Lung Association Research and Program Services Epidemiology and Statistics Unit. Trends in Pneumonia and Influenza Morbidity and Mortality; 2010. [3] Centers for Disease and Control. Updated recommendations for prevention of invasive pneumococcal disease among adults using the 23-valent pneumococcal polysaccharide vaccine (PPSV23). MMWR Morbidity and Mortality Weekly Report 2010:1102–6.
[4] Moberley S, Holden J, Tatham DP, Andrews RM. Vaccines for preventing pneumococcal infection in adults. The Cochrane Database of Systematic Reviews 2013;1:79 (review). [5] Vila-Corcoles A, Ochoa-Gondar O, Rodriguez-Blanco T, Raga-Luria X, GomezBertomeu F, EPIVAC Study Group. Epidemiology of community-acquired pneumonia in older adults: a population-based study. Respiratory Medicine 2009;103:309–16. [6] Welte T, Torres A, Nathwani D. Clinical and economic burden of communityacquired pneumonia among adults in Europe. Thorax 2012;67:71–9. [7] Colice GL, Morley MA, Asche C, Birnbaum HG. Treatment costs of communityacquired pneumonia in an employed population. Chest 2004;125:2140–5. [8] Carriere KC, Jin Y, Marrie TJ, Predy G, Johnson DH. Outcomes and costs among seniors requiring hospitalization for community-acquired pneumonia in Alberta. Journal of American Geriatric Society 2004;52:31–8. [9] Woodhead M, Blasi F, Ewig S, Garau J, Huchon G, Leven M, et al. Guidelines for the management of adult lower respiratory tract infections – full version. Clinical Microbiology and Infection 2011;17(Suppl 6):E1–59. [10] Huss A, Scott P, Stuck AE, Trotter C, Egger M. Efficacy of pneumococcal vaccination in adults: a meta-analysis. Canadian Medical Association Journal 2009;180(1):48–58. [11] UK Department of Health, Joint Committee on Vaccination and Immunisation. JCVI statement on discontinuation of the routine pneumococcal vaccination programme for adults aged 65 years and older, 16 March 2011. [12] Jackson LA, Janoff EN. Pneumococcal vaccination of elderly adults: new paradigms for protection. Clinical Infectious Diseases 2008;47:1328–38. [13] Klemets P, Lyytikäinen O, Ruutu P, Ollgren J, Nuorti P. Invasive pneumococcal infections among persons with and without underlying medical conditions: implications for prevention strategies. BMC Infectious Diseases 2008;8:96. [14] Food and Drug Administration. FDA expands use of Prevnar 13 vaccine for people ages 50 and older. Silver Spring, MD: US Department of Health and Human Services, Food and Drug Administration; December 2011. Available from http://www.fda.gov/newsevents/newsroom/pressannouncements/ucm 285431.htm [accessed 24.09.13]. [15] European Centre for Disease Prevention and Control. Technical Report of the Scientific Panel on Vaccines and Immunisation: use of pneumococcal polysaccharide vaccine for subjects over 65 years of age during an inter-pandemic period, Stockholm; 2007. [16] Kuehn BM. 10-year national vaccine plan aims to harness new technology, enhance safety. Journal of American Medical Association 2011;305(14):1399–400. [17] Salisbury DM, Beverley PCL, Miller E. Vaccine programmes and policies. British Medical Bulletin 2002;62:201–11. [18] Luchitsky A, Romanyuk G. An overview of GEOVAC: a software application to monitor immunization performance in Georgia, second version, For: Partners for Health ReformPlus; 2004. [19] Simons E, Mort M, Dabbagh A, Strebel P, Wolfson L. Strategic planning for measles control: using data to inform optimal vaccination strategies. The Journal of Infectious Diseases 2011;204:S28–34. [20] Ministry of Health in Israel, Public Health Services, Department of Epidemiology, 1999 vaccines briefing 2012 update. [21] The Johns Hopkins ACG® System. Technical reference guide 2009, version 9.0; December 2009. [22] Starfield B, Kinder K. Multimorbidity and its measurement. Health Policy 2011;103:3–8. [23] Evers SMAA, Ament AJHA, Colombo GL, Konradsen HB, Reinert RR, Sauerland D, et al. Cost-effectiveness of pneumococcal vaccination for prevention of invasive pneumococcal disease in the elderly: an update for 10 Western European countries. European Journal of Clinical Microbiology and Infectious Diseases 2007;26:531–40. [24] Pebody RG, Leino T, Nohynek H, Hellenbrand W, Salmaso S, Ruutu P. Pneumococcal vaccination Policy in Europe. Eurosurveillance 2005;10(7–9):174–8. [25] Centers for Disease Control. Prevention of pneumococcal disease: recommendations of the Advisory Committee on Immunization Practices. MMWR Morbidity and Mortality Weekly Report 1997;46(RR-8). [26] Shibli F, Chazan B, Nitzan O, Flatau E, Edelstein H, Blondheim O, et al. Etiology of community-acquired pneumonia in hospitalized patients in Northern Israel. Israel Medical Association Journal 2010;12:477–82. [27] Almirall J, Bolibar I, Serra-Prat M, Roig J, Hospital I, Carandell E, et al. New evidence of risk factors for community-acquired pneumonia: a populationbased study. European Respiratory Journal 2008;31:1274–84.
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
ARTICLE IN PRESS Vaccine xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling Ran D. Balicer a,b,∗ , Chandra J. Cohen a , Morton Leibowitz a , Becca S. Feldman a , Ilan Brufman a , Craig Roberts c , Moshe Hoshen a a
Clalit Research Institute, Chief Physician’s Office, Clalit Health Services, Arlozorov 101, Tel Aviv, Israel Department of Epidemiology, Faculty of Health Sciences, Ben Gurion University, Be’er Sheva, Israel c Pfizer Inc., Outcomes Research, 500 Arcola Road, Collegeville, PA 19301, USA b
a r t i c l e
i n f o
Article history: Received 11 March 2013 Received in revised form 12 November 2013 Accepted 10 December 2013 Available online xxx Keywords: Pneumococcal disease Pneumonia Vaccine strategy Risk profiling
a b s t r a c t Background: Current pneumococcal vaccine campaigns take a broad, primarily age-based approach to immunization targeting, overlooking many clinical and administrative considerations necessary in disease prevention and resource planning for specific patient populations. We aim to demonstrate the utility of a population-specific predictive model for hospital-treated pneumonia to direct effective vaccine targeting. Methods: Data was extracted for 1,053,435 members of an Israeli HMO, age 50 and older, during the study period 2008–2010. We developed and validated a logistic regression model to predict hospitaltreated pneumonia using training and test samples, including a set of standard and population-specific risk factors. The model’s predictive value was tested for prospectively identifying cases of pneumonia and invasive pneumococcal disease (IPD), and was compared to the existing international paradigm for patient immunization targeting. Results: In a multivariate regression, age, co-morbidity burden and previous pneumonia events were most strongly positively associated with hospital-treated pneumonia. The model predicting hospital-treated pneumonia yielded a c-statistic of 0.80. Utilizing the predictive model, the top 17% highest-risk within the study validation population were targeted to detect 54% of those members who were subsequently treated for hospitalized pneumonia in the follow up period. The high-risk population identified through this model included 46% of the follow-up year’s IPD cases, and 27% of community-treated pneumonia cases. These outcomes were compared with international guidelines for risk for pneumococcal diseases that accurately identified only 35% of hospitalized pneumonia, 41% of IPD cases and 21% of communitytreated pneumonia. Conclusions: We demonstrate that a customized model for vaccine targeting performs better than international guidelines, and therefore, risk modeling may allow for more precise vaccine targeting and resource allocation than current national and international guidelines. Health care managers and policymakers may consider the strategic potential of utilizing clinical and administrative databases for creating population-specific risk prediction models to inform vaccination campaigns. © 2013 Elsevier Ltd. All rights reserved.
1. Introduction Pneumococcal disease is responsible for a heavy burden of disease and mortality globally [1,2]. Invasive pneumococcal disease (IPD) is a critical and often fatal expression of pneumococcal
∗ Corresponding author at: Clalit Research Institute, Chief Physician’s Office, Clalit Health Services, Arlozorov 101, Tel Aviv, Israel. Tel.: +972 36923104; fax: +972 36925821. E-mail addresses: [email protected] (R.D. Balicer), [email protected] (C.J. Cohen), [email protected] (B.S. Feldman), [email protected] (I. Brufman), Craig.Roberts@pfizer.com (C. Roberts), [email protected] (M. Hoshen).
diseases, albeit relatively rare [3,4]. Pneumococcal pneumonias are more common and represent substantial costs from hospitalization and morbidity [5–8]. It is, however, difficult to identify the causative agent of pneumonia and to pinpoint those cases triggered by Streptococcus pneumonia [8]. Consequently, preventive strategies generally take a broad stroke approach to targeting risk for pneumonia, even though international guidelines issued by the Centers for Disease Control and Prevention’s Advisory Committee on Immunization Practices (ACIP) for identifying high-risk individuals have been based on research that identified risk of IPD [4,10]. While vaccines are considered a first line of defense against pneumococcal disease, the widely-used pneumococcal polysaccharide vaccine (PPSV23) is currently under scrutiny for its clinical
0264-410X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.vaccine.2013.12.020
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
2
effectiveness [1,11–13]. Much of the evidence in the literature supports PPSV23’s protective efficacy against IPD [1,3,10,12]; yet, evidence on its effectiveness in preventing pneumonias and other pneumococcal infections yields inconclusive results [4]. In parallel, a conjugate pneumococcal vaccine, considerably more costly, is an emerging alternative for pneumococcal disease prevention in adults. With the clinical efficacy of pneumococcal vaccines in question for preventing all-cause pneumonia and the FDA’s recent approval of a conjugate vaccine for use in older adults, aged 55 and up [14], policy-makers and health care managers are facing decisions about the extent of coverage and appropriate target populations for pneumococcal vaccines. Current national and international guidelines for administering pneumococcal vaccines are limited, usually based on broad age, chronic disease and immune-suppression criteria, yielding large, non-specific target groups [9,15]. While these universal clinical groupings are a valid starting point for vaccine targeting strategies, a more discriminative identification system is desirable to allow for effective resource allocation and to reach the most appropriate sub-groups at risk. The widespread integration of electronic medical records and large health information technology (IT) databases is being harnessed to support vaccine decision making and planning efforts [16]. Health IT has been incorporated into vaccine planning for aims such as: the identification of vaccine-attributable risk [17], immunization surveillance [18], and monitoring vaccine effectiveness [19]. Clalit Health Services (Clalit), the largest health fund in Israel, has the capacity to employ its electronic medical records database with integrated administrative and clinical (hospital and community-based) medical records to achieve more targeted vaccine planning. The health fund currently uses broadly-defined risk groups based on age and ACIP recommendations to target pneumococcal vaccine administration, similar to many health organizations worldwide. We propose that building a customized predictive model to identify sub-populations at risk for acquiring pneumococcal disease is an innovative approach to augmenting vaccine management. We present in this paper the development and validation of a population-specific predictive model for hospitaltreated pneumonia utilizing Clalit’s integrated longitudinal medical records data, and test the model in predicting invasive pneumococcal disease and community-treated pneumonia. 2. Materials and methods 2.1. Setting In Israel, membership in one of four nationwide health funds is universal and mandatory. Clalit is the largest of these funds, covering 53% of Israel’s population (4 million members) and serves as both an insurer and a health care provider supplying and financing services to members within its system. We used Clalit’s comprehensive database which is founded on both administrative claims data and clinical electronic health records, including data from laboratories, community clinics, hospitals, imaging centers, and pharmacies. Data from all medical encounters, including those in Clalit-owned hospitals and clinics, as well as those outside the Clalit system, are captured in the electronic medical records and collated in the organization’s centralized database. Laboratory and medications data are downloaded directly from the electronic database. 2.2. Study population and data collection This was a cohort study during the period January 1, 2008–December 31, 2010 using historical data collected from the Clalit Health Services’ centralized electronic database on all Clalit members age 50 and older (as of January 1, 2008). Those
members who switched funds (<1%) were excluded. The study period was divided into a baseline period (January 1, 2008–December 31, 2008), in which patient characteristics were recorded, and a follow up period from January 1, 2009–December 31, 2010 in which outcomes were documented. 2.3. Variables Potential risk factors from January 1, 2008 were incorporated into the model including: area-level socio-economic status (SES), ethnicity (defined by the patient’s designated primary care clinic), age, and gender. Clinical input variables comprised hospital-treated pneumonia events and community-treated pneumonia events in 2008, and smoking status (earliest available during the study period, with indicators more extensively available from 2009 onwards). To account for immunosuppression and susceptibility to pneumonia, disease-specific risk categories for pneumococcal immunization developed by the ACIP [3] and adapted by Clalit were included. This aggregate guideline-based categorization classifies patients into one of three groups: (1) low risk: Immunocompetent subjects without any major chronic medical conditions; (2) moderate risk: Immunocompetent with one or more chronic conditions of the heart, lung, liver and metabolism; (3) high risk: Immunocompromised patients including those with hematological conditions, functional or anatomic asplenia (patients recommended by ACIP to receive 2 vaccine doses) [20]. This classification system is based on the identification of chronic conditions that are either marked by impaired immune-competence or are known to predispose to pneumococcal disease. A complete list of these chronic diseases can be found in Supplement 1. We also integrated the Johns Hopkins Adjusted Clinical Groups (ACG) co-morbidity score into the model to measure co-morbidity burden at the individual patient level [21]. The ACG system is a well-tested and validated case-mix instrument for measuring patient complexity through the classification of co-morbidities based on assumed use of resources and costs. It categorizes morbidity by grouping individuals according to their age, gender and all known medical diagnoses (assigned over a defined period of time, typically one year) [22]. To facilitate modeling, the ACG System automatically assigns a sixlevel morbidity categorization termed Resource Utilization Bands (RUBs) [21]. 2.4. Outcomes Outcomes were episodes of hospital-treated all-cause pneumonia (HTP), community-treated all-cause pneumonia (CTP) and episodes of invasive pneumococcal disease. These events were identified through clinical and administrative records, from hospital diagnoses (ICD-9 coding), laboratory records (IPD cases based on positive microbiological blood tests taken from hospital laboratory records), and community health records (ICD-9 coding). Outpatient diagnoses made within 30 days of hospitalization or of previous outpatient diagnosis were counted as part of the previous hospitalized/non-hospitalized event, respectively. If more than 30 days after hospitalization the outpatient diagnosis was counted as a separate event. If an event included inpatient and outpatient diagnoses of pneumonia, it was defined as HTP. 2.5. Analysis Univariate analyses were performed to determine the independent risk factors that are associated with HTP incidence in the outcome period (2009–2010). Multivariate logistic regression was conducted to develop an adjusted model for HTP. The population age 50+ was randomly split in half for training and validating the logistic regression model. This model using 2008 baseline data
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
3
Table 1 Characteristics (as of 1 January 2008, unless otherwise stated) and relative risk for hospital-treated pneumonia (HTP) of total Clalit study population and of patients diagnosed with HTP during 2009–2010.
N Age Age group
Sex Ethnic group
Low SES indicatora New immigrant status Smoking status (as of end of 2009)
ACG morbidity burden – RUB (2008)
Pneumococcal disease risk group (ACIP-based guidelines, 2008) Previous pneumonia event (in 2008)
Mean 50–64 65–74 75–84 85+ Female Male General+ Orthodox Jewish Arab No Yes No Yes Never Past smoker Current smoker Unknown 0 1 2 3 4 5 Baseline Moderateb Highc No HTP HTP No CTP CTP
Entire Clalit population (50+ at onset of study 1.1.2008)
Members that had an HTP event in 2009–2010 (among members 50+ on 1.1.2008)
Relative risk (RR)
1,053,435 65.5 (SD = 11.2) 559,919 (53.2%) 242,634 (23.0%) 185,379 (17.6%) 65,503 (6.2%) 578,021 (54.9%) 475,414 (45.1%) 943,829 (89.6%)
20,971 75.83 (SD = 11.19) 3825 (18.2%) 4484 (21.4%) 7819 (37.3%) 4843 (23.1%) 10,087 (48.1%) 10,884 (51.9%) 18,728 (89.3%)
– – Reference 2.71 6.17 10.82 Reference 1.31 Reference
109,606 (10.4%) 772,161 (73.3%) 281,274 (26.7%) 329,632 (31.3%) 723,803 (68.7%) 711,616 (67.6%) 138,653 (13.2%) 160,360 (15.2%) 42,806 (4.0%) 28,459 (2.7%) 89,983 (8.5%) 134,878 (12.8%) 586,177 (55.6%) 153,972 (14.6%) 59,966 (5.7%) 515,884 (49.0%) 361,020 (34.3%) 176,531 (16.8%) 1,043,780 (99.1%) 9,655 (0.9%) 1,048,401 (99.5%) 5,034 (0.5%)
2243 (10.7%) 12,107 (57.7%) 8864 (42.3%) 2912 (13.9%) 18,059 (86.1%) 13,653 (65.1%) 3443 (16.4%) 2637 (12.6%) 1238 (5.9%) 157 (0.7%) 1360 (6.5%) 1236 (5.9%) 8945 (42.7%) 5272 (25.1%) 4001 (19.1%) 3596 (17.1%) 9977 (47.6%) 7398 (35.3%) 19,674 (93.8%) 1297 (6.2%) 20,704 (98.5%) 267 (1.3%)
1.03 Reference 2.01 Reference 2.82 Reference 1.29 0.86 1.51 Reference 2.74 1.66 2.77 6.21 12.09 Reference 3.96 6.01 Reference 7.13 Reference 2.69
95% CI
– – 2.59–2.82 5.94–6.42 10.38–11.28 1.28–1.35
0.99–1.08 1.96–2.07 2.72–2.94 1.25–1.34 0.82–0.89 1.42–1.60 2.32–3.23 1.41–1.96 2.36–3.24 5.30–7.27 10.32–14.18 3.82–4.12 5.78–6.25 6.76–7.51 2.39–3.02
a
Exempt from the national social security contribution due to low income. b Moderate risk group includes the following chronic conditions: cardiomyopathy, heart failure, ischemic heart disease, IHSS, vavular heart disease, carotid artery disease and CVA, bronchiectasis, chronic bronchitis, COPD, cystic fibrosis and history of tuberculosis. c High risk group includes the following chronic conditions or clinical indicators: transplant, malignancy, hodgkins, lymphoma, splenic dysfunction/splenectomy, chronic renal failure, dialysis.
to predict HTP was then tested for its sensitivity in predicting incidence of HTP, IPD and CTP during 2009–2010 in a validation population. The c-statistic and sensitivity of the model were used to measure model performance for the three outcomes. The predictive ability of the Clalit model was then compared to that of the high and moderate ACIP-based risk groups to identify HTP, IPD, and CTP. To compare the accuracy of the Clalit model in identifying those members who would acquire pneumonia or IPD during the outcome period to the ACIP-based risk category criteria, we used a point in the Clalit model distribution that corresponds to the equivalent population size of the ACIP-based risk groups, which are referred to in Table 3 as vaccine strategies: ACIP-based moderate risk group comprised 51% of the population (strategy 2), and therefore, the corresponding cut point in the Clalit risk model was 51% (strategy 3). Whereas, the ACIP-based high risk group comprised only 16.8% (∼17%) of the total study population (strategy 4), therefore an equivalent proportion of the highest risk was taken at the cut point of 17% in the Clalit model (strategy 5). In evaluating different vaccine targeting strategies, we also calculated what percentage of the population would have to be targeted to yield the same 35% sensitivity as using the ACIP-based criteria (strategy 6). A more extreme cut point of 5% was also tested (strategy 7), to represent a strategy reflecting very limited available resources. The predictive measures were tested for statistical significance using the Chi-square test. SPSS V 18.0 was used for analysis. The institutional ethics committee at Clalit approved this study.
3. Results This population-specific study of risk for pneumonia and IPD is based on an analysis of data garnered from the records of 1,053,435 patients over the age of 50 as of January 1, 2008. The study population characteristics and relative risk (RR) assessments are presented in Table 1. The study population was 54.9% female, with the following age group distribution: 53.2% were 50–64 years old, 23.0% were 65–74 years old, 17.6% were 75–84 years old, and 6.2% were 85 and older. More than a quarter of the population (26.7%) was low income status, 28.4% were current or past smokers, and 10.4% were patients residing in predominantly Arab localities. Of the adult population age 50 and older, 51% met one or more of the criteria for high or moderate ACIP-based risk groups, while 36% of those age 50 through 64 met one or more of the ACIP criteria. Supplement 1 details the prevalence of each chronic condition in the study population. There were 20,917 members (2.0% of the Clalit population studied) that were identified as having had at least one hospital-treated pneumonia event in 2009–2010 (Table 1). Those who had a hospitalized pneumonia event were on average 75.8 ± 11.2 years old, 29% were past or current smokers, 10.7% were patients from Arab localities, and 42.3% were patients residing in areas of low SES. Table 1 shows that risk for hospitalized pneumonia among Clalit members increased with age: members age 65–74 had a relative risk (RR) of 2.71 (95% CI = [2.59–2.82]) compared to members 50–64; those ages 75–84 had an RR of 6.17 (95%
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
4
Table 2 Multivariate regression analysis of hospitalized pneumonia risk from 2009-2010 among patients 50 years and older in Clalit (training population). Mulitvariate Analysis n = 526,718
Sig.
Days hospitalized (2008)
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.007 0.844 0.249 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Age group Sex Ethnic group Low SES indicator New immigrant status Smoking status (2010)
ACG morbidity burden – RUB (2008)
Pneumococcal disease risk group (ACIP-based MoH definitions, 2008) Previous pneumonia event (2008)
50–64 65–74 75–84 85+ Male Arab Yes Yes Never Past smoker Current smoker Baseline 1 2 3 4 5 Baseline Moderate High Hospital-treated event Community-treated event
B
Exp(B) [OR]
95% C.I. for OR Lower
Upper
0.012
1.012
1.010
1.014
0.598 1.308 1.872 0.247 0.256 0.364 0.241
1.818 3.697 6.501 1.280 1.292 1.439 1.273
1.704 3.472 6.055 1.226 1.211 1.380 1.197
1.940 3.936 6.980 1.336 1.379 1.502 1.353
0.141 0.318
1.151 1.374
1.087 1.290
1.219 1.464
0.381 −0.028 0.156 0.485 0.782
1.464 0.973 1.169 1.625 2.187
1.110 0.738 0.896 1.243 1.670
1.929 1.281 1.523 2.123 2.863
0.761 0.915 0.943 0.583
2.141 2.496 2.568 1.792
2.015 2.335 2.334 1.488
2.275 2.669 2.827 2.159
Table 3 Strategies for vaccine targeting (test population). Vaccination strategy 1 2 3 4 5 6 7 a b c
High and Moderate risk groups and all aged 65+ High and Moderate ACIP-based risk groups Clalit model, 51% highest risk scores (based on training model) ACIP-based highest risk group (Immunosuppressed) Clalit model, 17% highest risk scores (based on training model) Clalit model, 8.6% highest risk scores (based on training model) Clalit model, 5% highest risk scores (based on training model)
% of HTP cases in 2009–2010 identified (n = 10,423)
% of IPD cases in 2009-10 identified (n = 90)
66% (347,008)
94% (9818)
89% (80)
78% (3572)
51% (268,616)
a
83% (8609)
80% (72)
65% (2980)b
51% (267,744)
85% (8896)a
80% (72)
66% (3045)b
17% (88,142)
35% (3634)c
41% (37)
21% (971)c
17% (87,853)
54% (5667)c
46% (41)
27% (1246)c
8.6% (45,521)
35% (3634)
31% (28)
15% (692)
5% (25,580)
23% (2390)
18% (16)
9% (421)
% of 50+ population targeted (n = 526,717)
% of CTP cases in 2009–2010 identified (n = 4,603)
When comparing strategies 2 and 3, p < 0.001 using McNemar test. When comparing strategies 2 and 3, p < 0.05 using McNemar test. When comparing strategies 4 and 5, p < 0.001 using McNemar test.
CI = [5.94–6.42]), and those age 85 and older had an RR of 10.82 (95% CI = [10.38–11.28]). Male gender RR = 1.31; [CI = 1.28–1.35], and morbidity burden, as measured by the ACG RUBs, had RRs reaching 12.09 [CI = 10.32–14.18] for members in the highest morbidity burden group. The risk of HTP was higher among members with low socio-economic status (SES), with an RR = 2.01 (95% CI = 1.96–2.07). In Table 2 we present the results of the multivariate logistic regression model. HTP is positively associated with age, those age 75–84 had an OR = 3.70 and those age 85+ had an OR = 6.50 (95% CI = 6.06–6.98) compared to those age 50–64. A recent history of hospital-treated or community-treated pneumonia increased the risk of another event with ORs of 2.57 (95% CI = 2.33–2.82) and 1.79 (95% CI = 2.33–2.82), respectively. ACIP-based risk groups were associated with significantly increased risk beyond many of the above risk factors (OR = 2.50 [95% CI = 2.34–2.67]) for the high risk group and OR = 2.14 [95% CI = 2.02–2.28] for the moderate risk group), and a high morbidity burden RUB status had an additional significant effect (OR = 1.63 [95% CI = 1.24–2.12] and 2.19
[95% CI = 1.67–2.86] for the highest 2 categories). While current smoking appeared to be protective for HTP in the univariate model, (RR of 0.86; 95% CI = 0.82–0.89), in the multivariate model, adjusted for multiple confounders, it becomes a significant risk factor for HTP (RR = 1.37; 95% CI = 1.29–1.47) and past smoker is a weaker yet significant risk factor (RR = 1.15; 95% CI = 1.09–1.22) compared to members who never smoked. The performance indicators for the Clalit tailored model were high for both the training and test populations, with a c-statistic of 0.80 for the test population model. In Table 3 we illustrate several vaccine targeting strategy options and the application of the HTP calibrated model to predict other related clinical syndromes, IPD and CTP. With a narrow target population of 17% (the percentage that corresponds to the ACIP-based high-risk category within the study population), sensitivity comparisons show that the ACIP-based strategy 4 yielded 35%, 41% and 21% sensitivity predicting HTP, IPD and CTP, respectively; while the Clalit tailored model strategy 5 yielded 54%, 46% and 27%, for each of the three
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
outcomes, respectively. Comparison of sensitivities measured were statistically significant for both HTP and CTP (p < 0.001) and was not statistically significant for IPD, given low incidence (p = 0.222). Looking at a broader target population, the ACIP-based strategy 2, these high/moderate risk patients comprised 51% of the entire population, and 83% of the HTP cases occurred in this population segment. Selecting 51% of the population, strategy 3, using our customized model yielded a slightly higher sensitivity of 85%, with differences between these two strategies statistically significant at p < 0.001 for HTP. In an additional analysis, predicting and identifying 35% of the HTP cases with our model would involve targeting vaccination for only 8.6% of adults age 50 and older, as compared to 17% when using ACIP-based criteria.
4. Discussion Immunization programs in many countries in Europe and North America have endorsed universal age-based pneumococcal vaccination policies among older adults (over the age of 65), as well as for immune-compromised younger adults [9,15,23,24]. In the United States, these recommendations by the ACIP are longstanding for over a decade [25]. This widespread strategy is, however, being brought into question even in countries where the vaccines were routinely administered [11]. When applied in practice, many patients, whose risk for pneumonia or IPD is modest at most, get vaccinated based solely on age. As more costly vaccines become widely available, health provider organizations in Israel and in other countries may seek to further target populations for vaccination campaigns. Our findings suggest that country-level guidelines can be tailored for an organization’s specific population characteristics, epidemiologic circumstances, and financial structures to optimize allocation of limited resources of a subsidized vaccination outreach program. In this study we document an initial effort to utilize an integrated electronic medical database for population-specific targeting in pneumococcal vaccine management to optimize disease prevention and resource planning. Risk profiles for pneumococcal disease have predominantly been supported by research focused on IPD, as it is the one illness event in which pneumococcus can definitively be identified as the causative organism. It has also been shown, however, that a large proportion of hospital treated pneumonia is caused by pneumococcus (Streptococcus pneumonia) [9,26]. Therefore, we examined risk factors for HTP in our model because the costs and burden of hospitalized pneumonias are a significant concern for pneumococcal disease prevention strategies [2,5,6]. To build a predictive model customized to Clalit’s population, we combined widely-accepted risk parameters such as age, sex, smoking status and socio-economic status (SES) with a national ACIP-based risk classification, and patient morbidity burden as measured by John Hopkins Adjusted Clinical Groups (ACG) based RUB index. This customized model elucidates the following as the most prominent risk factors for HTP among Clalit members: older age (75+), high morbidity (from both the ACG morbidity and ACIP-based risk classifications), and having experienced a previous pneumonia event. While age and previous pneumonia events have been identified in the literature as important risk factors, the use of the ACG morbidity burden indicator provides an important additional prognostic indicator beyond immunosuppression to strengthen our predictive model. Accurately predicting those individuals who will experience a future pneumonia event to enable preventative action underscores the motivation for building this model. Among Clalit’s total population 50 and older, 20,971 (2%) individuals were identified as having an HTP event, 192 having an IPD event, and 9250 having a CTP event in 2009–2010. When compared in an equivalent proportion
5
of the population (17%), the Clalit custom model exhibited higher predictive utility for HTP than using ACIP-based risk profiles, as shown in Table 3 (comparison of strategies 4 and 5). Additionally, the Clalit model would allow more judicious use of vaccines to capture the same 35% of HTP cases, targeting half as many of its 50+ adult population (8.6% vs. 17%) than if it were using ACIP-based risk criteria (comparing strategies 4 and 6). When comparing strategies toward a broader target population (strategies 2 and 3), the differences in sensitivities are statistically significant but of minimal population impact. A customized model is thus clinically relevant only when resources are considerably limited, and do not allow for vaccination of the entire high/moderate risk group determined by the ACIP criteria. The added value of using the customized model becomes apparent when greater targeting is required. Through our evaluation we have demonstrated the ability of an organization to use its database to develop a risk model with better predictive accuracy than internationally-defined criteria to effectively identify high-risk sub-groups of patients for vaccine administration targeting. 4.1. Limitations Our study has some important limitations to take into consideration. While we included a large set of risk factor variables in our logistic regression analysis, we did not incorporate an exhaustive list of factors. Previous studies on risk factor identification for pneumococcal disease have identified additional key variables to those that were part of our study, including: medications, frailty indices, allergies, pets, environmental irritants, and poor oral hygiene [27]. The predictive accuracy of our model as measured by the c-statistic 0.80 is considered robust; however with the incorporation of additional risk variables, the model can potentially provide greater predictive value. Using ecological clinical variables such as SES and ethnicity may have caused some confounding bias, thereby reducing the sensitivity of our model. 4.2. Implications for practice Utilizing this customized risk model developed within Clalit provides a predictive advantage over international guideline risk criteria, for the purpose of targeting pneumococcal vaccines. As illustrated in Table 3, specific targets for vaccination can be tailored to better balance clinical risks associated with vaccine administration. In Clalit’s population, more than 1 million members (25% of all members) are 50 years and older and almost 500.000 members are age 65 and older. In our test population (n = 5,26,717), when considering both age-based (65+) and clinical risk groupings currently recommended by the ACIP, 66% (3,47,008) of Clalit members aged 50+ meet the vaccination criteria. Ignoring the agebased recommendations and employing the ACIP clinical high and moderate-risk criteria, still over 50% of members (268,616) aged 50+ meet one or more of these criteria. Targeting only the highest ACIP-based risk group for pneumococcal disease (comprised of immune-suppressed members) for administration would entail providing vaccines for 88,142 Clalit members (17%) – potentially still a large target for a costly vaccination campaign. If this target population size was not manageable for the organizational resources available, the model could be adjusted to use a lower cut point, such as 5% (strategy 7 in Table 3). As a function of the power of a customized strategy, the levels of disease prevention remain high despite a substantial reduction in the target population size. Using clinical and administrative data to direct vaccination policy helps to prioritize organizational planning, resource allocation and clinical decision-making through the identification of key highrisk sub-groups and the monitoring of vaccination administration.
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020
G Model JVAC-14933; No. of Pages 6
ARTICLE IN PRESS R.D. Balicer et al. / Vaccine xxx (2014) xxx–xxx
6
With electronic patient health records (EHRs), an organization can identify the 50+ cohort and determine risk profiles for organizationspecific vaccination guidelines. A diverse set of risk factors can be tested and added to the model, depending on availability of data. These risk factors can be integrated into an automated algorithm within the EHR database to flag patients at high risk, for practitioners accessing patients’ files or to create lists of candidates for vaccination outreach. Using our methodology, a risk cut-off threshold can be adjusted to accommodate the budget and operational needs of the organization or physician practice. 5. Conclusion A customized risk model has the potential to enhance cost management and facilitate clinical efforts by focusing costs on the individuals in the population who stand to benefit most, and targeting future vaccines to patients at highest risk. We expect that such predictive modeling will become widespread in many other clinical applications as well. We recommend that health care managers and policy-makers with access to clinical and administrative databases explore the potential for creating population-specific risk prediction models for more effective vaccination targeting. Funding This study was funded in part by an unconditional institutional research grant from Pfizer Inc. The authors did not receive any individual financial contribution from this funding source. The sponsor provided methodological comments for the study design, and critically reviewed outcome reports and the manuscript. Conflicts of interest C.R. is an employee of Pfizer Inc., and has served as a methodological adviser to the research group. None of the other authors have any conflicts of interest to declare. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.vaccine. 2013.12.020. References [1] Moberley S, Holden J, Tatham DP, Andrews RM. Vaccines for preventing pneumococcal infection in adults. The Cochrane Database of Systematic Reviews 2008;4:57 (review). [2] American Lung Association Research and Program Services Epidemiology and Statistics Unit. Trends in Pneumonia and Influenza Morbidity and Mortality; 2010. [3] Centers for Disease and Control. Updated recommendations for prevention of invasive pneumococcal disease among adults using the 23-valent pneumococcal polysaccharide vaccine (PPSV23). MMWR Morbidity and Mortality Weekly Report 2010:1102–6.
[4] Moberley S, Holden J, Tatham DP, Andrews RM. Vaccines for preventing pneumococcal infection in adults. The Cochrane Database of Systematic Reviews 2013;1:79 (review). [5] Vila-Corcoles A, Ochoa-Gondar O, Rodriguez-Blanco T, Raga-Luria X, GomezBertomeu F, EPIVAC Study Group. Epidemiology of community-acquired pneumonia in older adults: a population-based study. Respiratory Medicine 2009;103:309–16. [6] Welte T, Torres A, Nathwani D. Clinical and economic burden of communityacquired pneumonia among adults in Europe. Thorax 2012;67:71–9. [7] Colice GL, Morley MA, Asche C, Birnbaum HG. Treatment costs of communityacquired pneumonia in an employed population. Chest 2004;125:2140–5. [8] Carriere KC, Jin Y, Marrie TJ, Predy G, Johnson DH. Outcomes and costs among seniors requiring hospitalization for community-acquired pneumonia in Alberta. Journal of American Geriatric Society 2004;52:31–8. [9] Woodhead M, Blasi F, Ewig S, Garau J, Huchon G, Leven M, et al. Guidelines for the management of adult lower respiratory tract infections – full version. Clinical Microbiology and Infection 2011;17(Suppl 6):E1–59. [10] Huss A, Scott P, Stuck AE, Trotter C, Egger M. Efficacy of pneumococcal vaccination in adults: a meta-analysis. Canadian Medical Association Journal 2009;180(1):48–58. [11] UK Department of Health, Joint Committee on Vaccination and Immunisation. JCVI statement on discontinuation of the routine pneumococcal vaccination programme for adults aged 65 years and older, 16 March 2011. [12] Jackson LA, Janoff EN. Pneumococcal vaccination of elderly adults: new paradigms for protection. Clinical Infectious Diseases 2008;47:1328–38. [13] Klemets P, Lyytikäinen O, Ruutu P, Ollgren J, Nuorti P. Invasive pneumococcal infections among persons with and without underlying medical conditions: implications for prevention strategies. BMC Infectious Diseases 2008;8:96. [14] Food and Drug Administration. FDA expands use of Prevnar 13 vaccine for people ages 50 and older. Silver Spring, MD: US Department of Health and Human Services, Food and Drug Administration; December 2011. Available from http://www.fda.gov/newsevents/newsroom/pressannouncements/ucm 285431.htm [accessed 24.09.13]. [15] European Centre for Disease Prevention and Control. Technical Report of the Scientific Panel on Vaccines and Immunisation: use of pneumococcal polysaccharide vaccine for subjects over 65 years of age during an inter-pandemic period, Stockholm; 2007. [16] Kuehn BM. 10-year national vaccine plan aims to harness new technology, enhance safety. Journal of American Medical Association 2011;305(14):1399–400. [17] Salisbury DM, Beverley PCL, Miller E. Vaccine programmes and policies. British Medical Bulletin 2002;62:201–11. [18] Luchitsky A, Romanyuk G. An overview of GEOVAC: a software application to monitor immunization performance in Georgia, second version, For: Partners for Health ReformPlus; 2004. [19] Simons E, Mort M, Dabbagh A, Strebel P, Wolfson L. Strategic planning for measles control: using data to inform optimal vaccination strategies. The Journal of Infectious Diseases 2011;204:S28–34. [20] Ministry of Health in Israel, Public Health Services, Department of Epidemiology, 1999 vaccines briefing 2012 update. [21] The Johns Hopkins ACG® System. Technical reference guide 2009, version 9.0; December 2009. [22] Starfield B, Kinder K. Multimorbidity and its measurement. Health Policy 2011;103:3–8. [23] Evers SMAA, Ament AJHA, Colombo GL, Konradsen HB, Reinert RR, Sauerland D, et al. Cost-effectiveness of pneumococcal vaccination for prevention of invasive pneumococcal disease in the elderly: an update for 10 Western European countries. European Journal of Clinical Microbiology and Infectious Diseases 2007;26:531–40. [24] Pebody RG, Leino T, Nohynek H, Hellenbrand W, Salmaso S, Ruutu P. Pneumococcal vaccination Policy in Europe. Eurosurveillance 2005;10(7–9):174–8. [25] Centers for Disease Control. Prevention of pneumococcal disease: recommendations of the Advisory Committee on Immunization Practices. MMWR Morbidity and Mortality Weekly Report 1997;46(RR-8). [26] Shibli F, Chazan B, Nitzan O, Flatau E, Edelstein H, Blondheim O, et al. Etiology of community-acquired pneumonia in hospitalized patients in Northern Israel. Israel Medical Association Journal 2010;12:477–82. [27] Almirall J, Bolibar I, Serra-Prat M, Roig J, Hospital I, Carandell E, et al. New evidence of risk factors for community-acquired pneumonia: a populationbased study. European Respiratory Journal 2008;31:1274–84.
Please cite this article in press as: Balicer RD, et al. Pneumococcal vaccine targeting strategy for older adults: Customized risk profiling. Vaccine (2014), http://dx.doi.org/10.1016/j.vaccine.2013.12.020