- Email: [email protected]
Chlamydia Screening and Pelvic Inflammatory Disease
Chlamydia Screening and Pelvic Inflammatory Disease Insights from Exploratory Time–Series Analyses Kwame Owusu-Edusei Jr, PhD, Michele K. Bohm, MPH, Harrell W. Chesson, PhD, Charlotte K. Kent, PhD Background: Screening for chlamydia has been reported to reduce pelvic inflammatory disease (PID) at the individual level. However, information on population-level association (or causality) is scant.
Purpose: This study aims to examine the association between chlamydia and gonorrhea screening and PID diagnoses using time–series analyses. Methods: Monthly chlamydia and gonorrhea screening and PID diagnosis rates were extracted for a cohort of 207,695 continuously enrolled privately insured women from January 2001 to December 2006. An autoregressive integrated moving average model was used to examine whether rates of PID diagnoses in a given month were associated with rates of chlamydia and gonorrhea screening in previous months.
Results: Monthly screening rates increased from about 300 to almost 700 per 100,000 for chlamydia and from 250 to almost 650 per 100,000 for gonorrhea, whereas PID diagnosis rates declined during the same period (40 –20 per 100,000). Increases in screening rates were associated with decreases in PID diagnosis rates 4 months later. On average, a one-unit (or 10%) increase in the growth of chlamydia and gonorrhea screening rates, separately, in the prior fourth month was signifıcantly associated with a 0.36 (or 3.6%, p⬍0.05) and 0.32 (or 3.2%, p⬍0.10) decrease in the growth rate of the PID diagnosis rate, respectively. Conclusions: Although analyses such as these cannot prove causality, the results are consistent with the hypothesis that increases in chlamydia and gonorrhea screening coverage can lead to reductions in PID at the population level. A population-level focus offers advantages over individual-level analyses of screening and PID, such as the ability to capture indirect benefıts of increased screening. (Am J Prev Med 2010;38(6):652– 657) Published by Elsevier Inc. on behalf of American Journal of Preventive Medicine
Introduction
P
elvic inflammatory disease (PID) has been broadly described1 as a range of upper genital tract disorders among women, including one or a combination of the following: endometritis, salpingitis, tubo-ovarian abscess, and pelvic peritonitis. PID is caused by the introduction of pathogenic micro-organisms from the lower to the upper reproductive organs.2 It develops in 10% to 40% of untreated chlamydial or gonococcal cervicitis cases in women.1 Thus, although gonorrhea (GC) and chlamydia are known common causes of PID, a substan-
From the Division of STD Prevention, CDC, Atlanta, Georgia Address correspondence to: Kwame Owusu-Edusei Jr, PhD, Division of STD Prevention, CDC, 1600 Clifton Road, MS E-80, Atlanta GA 30333. E-mail: [email protected]. 0749-3797/00/$17.00 doi: 10.1016/j.amepre.2010.02.008
652
Am J Prev Med 2010;38(6):652– 657
tial proportion of PID cases have other causes or no known etiology.1 A seminal RCT published in 1996 found that selective testing for chlamydia in at-risk women reduced the incidence of PID.3 Since then, two other studies4,5 have reported comparable results in different settings using different methods. However, potential limitations of these studies have been noted and the magnitude of the utility of chlamydia screening as a tool to prevent PID has subsequently been debated.6 –10 Past analyses11 of claims data have found that young women diagnosed with PID were more likely to have been previously screened for chlamydia than young women without a PID diagnosis. This positive association between chlamydia screening and PID (which seems at odds with the idea that chlamydia screening can reduce PID rates12) likely arises because (1) screening is targeted
Published by Elsevier Inc. on behalf of American Journal of Preventive Medicine
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
toward those with risk factors for chlamydia and (2) a notable portion of the screening tests might actually be diagnostic tests among those seeking care as a result of clinical symptoms.11 Because these important individuallevel confounders that influence the risk of PID cannot be controlled for in analyses using claims data, this study focused on changes in screening coverage rates and changes in PID diagnosis rates at the population level using time–series analyses. A population-level approach reduces the confounding of the association between screening and PID at the individual level and allows us to address the question of whether changes in the frequency of screening can affect PID rates at the population level. In addition, a population-level focus has the advantage of being able to capture indirect benefıts (such as interruption of disease transmission and reduced population prevalence) of chlamydia and gonorrhea screening and treatment on PID rates. One method used to investigate causality (or association) for time–series data is the Granger causality technique.13 According to Granger,14 if past measures of a variable/event (X) helps to predict another (Y), then X “Granger causes” Y. Following its application to economic data in 1972,13 several other fıelds, including medicine, have employed it to study relationships between data sets over time.15–20 The authors are unaware of the application of the Granger causality technique to study
653
sexually transmitted diseases and their sequelae. Application of this technique (i.e., Granger-causality) may potentially assist and/or complement results/analyses from public health surveillance data and other data sources such as claims data (particularly as the nation moves to richer and more extensive electronic medical record systems). In this study, the Granger causality technique was used to investigate potential population-level associations between monthly chlamydia/gonorrhea screening and PID diagnosis.
Methods
Outpatient and inpatient claims data from the MarketScan Database (MEDSTAT Group) for 2001 through 2006 were used. The database contains claims data on more than 5 million people who have employer-sponsored health insurance from more than 100 payers, including large employers, health plans, and government and public organizations. The employer-sponsored privately insured constitute about 58% of the population in the U.S. and analyses from the MarketScan Database can be projected to represent that population (Thompson Medstat Group). Monthly claims data were extracted for a cohort of 207,695 women (aged 15–39 years in 2001) who were enrolled continuously from January 1, 2001, through December 31, 2006 —72 months. Each of the women in the cohort was enrolled for a total of 2191 member-days (i.e., 365 days for each year except 2004, which was a leap year). Current procedural terminology (CPT) codes for chlamydia screening (87110, 87270, 87320, 87490, 87491, and 87810) and gonorrhea screening (87590, 87591, 87592, and 87850) were used to identify the number of women in the cohort who were screened Table 1. Number and distribution of PID diagnosis codes each month. The ICD-9 codes for chlamydia Code Description Frequencya Percentage (078.88, 079.88, 079.98, 099.41, and 099.50 – 098.10 Acute GC upper GU tract, site unspecified 7 0.04 099.59) and gonorrhea (098.0 – 098.89) diagnoses 098.16 Acute GC endometritis 15 0.09 were used to construct 098.17 Acute GC salpingitis 39 0.24 and examine monthly di098.19 Acute GC upper GU tract, other site 30 0.19 agnosis rates. National Drug Codes (NDC) of 098.86 Acute GC peritonitis 64 0.40 drugs recommended by 099.56 Acute chlamydia peritonitis 2 0.01 the CDC for treating chlamydia and gonorrhea21 614.0 Acute salpingo-oophoritis 656 4.10 from drug claims tables 614.2 Salpingitis/oophoritis, not acute or chronic 1950 12.20 were used to estimate treatment rates. 614.3 Acute parametritis/PID 1680 10.51 Women in the cohort 614.5 Acute or unspecified pelvic peritonitis 10 0.06 diagnosed with acute PID in each month were 614.8 Other specified inflammatory disease, female pelvic organs 119 0.74 identifıed using ICD-9 614.9 Unspecified inflammatory disease, female pelvic organs 8205 51.31 codes (Table 1) from 615.0 Inflammatory disease of uterus, except cervix 787 4.92 the inpatient and outpatient claims tables. 615.9 Unspecified inflammatory disease of uterus 2426 15.17 The time–series analyTotal 15,990 100 ses performed in this a study examined whether Note that in a substantial number of instances, more than one diagnosis code was associated with an changes in chlamydia and enrollee in 1 month. gonorrhea screening rates GC, gonorrhea; GU, genitourinary; PID, pelvic inflammatory disease
June 2010
654
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
Figure 1. Monthly chlamydia and gonorrhea screening rates (per 100,000) and pelvic inflammatory disease (PID) diagnosis rates (per 100,000) for a cohort of employersponsored privately insured women aged 15–39 years in 2001 who were continuously enrolled from January 2001 through December 2006. Note that the vertical axes have different scales.
creased substantially (from their lowest points in the early months of 2001) to the end of the period examined; chlamydia screening rates increased from about 300 to almost 700 (a 133% increase) and gonorrhea screening rates increased from about 250 to almost 650 (a 160% increase). The monthly PID diagnosis rate decreased from about 40 to 20 (a 50% decrease) during the same period. The screening rate for chlamydia was 4%–28% higher than the screening rate for gonorrhea in each month. Thus, for this cohort, chlamydia and gonorrhea screening rates trended upward, whereas the PID diagnosis rates trended downward (see Figure 1). In addition, the chlamydia and gonorrhea series had similar trends, with a signifıcant (p⬍0.001) high contemporaneous correlation coeffıcient of 0.97. Diagnosis rates for both chlamydia and gonorrhea increased for each subsequent year. For the cohort examined, the diagnosis rate (number per 100,000) for chlamydia increased from about 5 in January 2001 to about 13 in December 2006, and from 2 in January 2001 to about 6 in December 2006 for gonorrhea. However, the treatment rate for the cohort was notably low (⬍40%). Results of the regression from models 1, 2, and 3 are presented in Table 2. More detailed information on the time–series regression results has been provided in Appendix B (available online at www.ajpm-onlinenet). In both cases, the fourth lags (4-month lag between screening and PID diagnosis) were signifıcant (p⬍0.10), although the coeffıcient for chlamydia was signifıcant at 5% (see Table 2). The coeffıcients for the fourth lag of chlamydia indicated that a one-unit (or 10%) increase in the growth rate of chlamydia screening rate in the past fourth month was associated with a 0.36 (or 3.6%) decrease in the growth rate of PID diagnosis, on average. Similarly, a one-unit (or
in a given month were associated with changes in PID rates in subsequent months. See Appendix A (available online at www.ajpmonline.net) for details on the time–series analyses. The signifıcance level was set at 10% because of the relatively small number of observations (n⫽72 months). The analysis was repeated for two mutually exclusive age categories—women aged 15–24 years and 31–39 years in 2001. As a test against spurious correlations, examination of whether changes in screening coverage rates in the younger cohort were associated with changes in PID diagnosis rates in the older cohort, and whether changes in screening coverage rates in the older cohort were associated with changes in PID rates in the younger cohort, were conducted. The idea behind this additional test was that if screening is causally associated with subsequent reductions in PID diagnosis, then changes in screening coverage in the younger (or older) cohort would be expected to have much more of an impact on PID rates in that cohort than in the Table 2. Factors associated with monthly PID diagnosis rates: Time–series regression other cohort. The analyanalysis coefficients (t statistics) ses were conducted in 2008 and 2009 using Variable Model 1 Model 2 Model 3 SAS, version 9.2, and Eviews, version 6. Constant ⫺0.02 (⫺1.84)* ⫺0.01 (⫺1.99)* ⫺0.01 (⫺1.92)*
Results Figure 1 shows monthly screening rates (number screened per 100,000) for chlamydia and gonorrhea, and PID diagnosis rates (number of cases per 100,000). Chlamydia and gonorrhea screening rates in-
PID rate, month t-1
⫺0.92 (⫺7.8)***
⫺0.98 (⫺12.9)***
⫺0.99 (⫺12.6)***
PID rate, month t-2
⫺0.53 (⫺4.5)***
⫺0.52 (⫺8.6)***
⫺0.52 (⫺8.0)***
Chlamydia screening rate, month t-4
—
Gonorrhea screening rate, month t-4
—
Adjusted R2
0.53
⫺0.36 (⫺2.0)** — 0.65
— ⫺0.32 (⫺1.9)* 0.65
Note: Regression coefficients and t statistics (in parentheses) are shown. PID rates and screening rates for chlamydia and gonorrhea were expressed in terms of the change in the log value from the previous month. A complete version of this table is provided in Appendix B (available online at www.ajpm-online.net). *0.1, **0.05, ***0.01 PID, pelvic inflammatory disease
www.ajpm-online.net
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
655
0.28 (p⬍0.10) decrease in the growth rate of PID diagnosis for the younger and older age groups, respectively (results not shown). In testing the model for susceptibility to spurious associations, this study found that changes in chlamydia screening rates in the younger age cohort were not associated with changes in PID rates in the older age cohort, and changes in chlamydia screening rates in the older age group were not associated with changes in PID in the younger age group (p⬎0.3).
Discussion Figure 2. Monthly chlamydia screening rate (per 100,000) and pelvic inflammatory disease (PID) diagnosis rates (number per 100,000) for employer-sponsored privately insured women aged 15–24 years and 31–39 years in 2001 who were continuously enrolled from January 2001 through December 2006. Note that the vertical axes have different scales.
10%) increase in gonorrhea screening rate in the past fourth month was associated with a 0.32 (or 3.2%) decrease in the growth rate of PID diagnosis, on average. As an example, if the growth rate of chlamydia screening increased by one unit, say from 6% to 7% (representing an increase from 400/100,000 to 428/100,000, instead of to 424/100,000), one would expect the growth rate of PID diagnosis 4 months later to decrease by 0.36 from ⫺3% to ⫺3.36% (representing a decrease from 30/100,000 to 28.92/ 100,000, instead of to 29.1/100,000). Similarly, a one-unit increase (6% to 7%) in the growth rate of gonorrhea screening would be expected to decrease the growth rate of PID diagnosis 4 months later by 0.32 from ⫺3% to ⫺3.32% (representing a decrease from 30/100,000 to 29.04/100,000, instead of to 29.1/100,000). Monthly screening rates for chlamydia and PID diagnosis rates for the two age groups (women aged 15–24 years and 31–39 years in 2001) are shown in Figure 2. The screening rate for the older age group showed a slight upward trend but was substantially lower than the screening rate for the younger age group. The PID diagnosis trend for the older age group was also substantially lower than for younger women and was fairly stable, vacillating at approximately 15 cases per 100,000 women. Chlamydia screening and PID diagnosis rates for younger women were substantially higher and exhibit noticeably steeper trends and higher variance. As with the base case analysis, changes in chlamydia screening rates helped to predict changes in the subsequent PID diagnosis rates within each age group. On average, a one-unit increase in the growth rate of chlamydia screening was associated with a 0.38 (p⬍0.10) and June 2010
To the authors’ knowledge, this is the fırst application of time–series analysis of medical claims data to explore the potential impact of chlamydia and gonorrhea screening in terms of preventing PID. This study found that increases in the 4-month lag of both the chlamydia and gonorrhea screening series improved the predictive power of the PID model. That is, screening rates in a given month helped to predict PID rates 4 months later. Thus, the results suggest that there are population-level benefıts from increases in chlamydia/gonorrhea screening in terms of reducing PID. Although this observational study design cannot establish a causal association between increased screening and decreased PID, the results are consistent with a causal association as well as with previously published studies3–5 showing that increased chlamydia/ gonorrhea screening was associated with reductions in the burden of PID. The thesis that chlamydia screening may reduce PID is based on the idea that the cases identifıed would be treated, subsequently preventing PID. This study found an unrealistically low treatment rate for those diagnosed with chlamydia and/or gonorrhea after screening. The low treatment rate may largely be due to one (or a combination) of these reasons: treatment with free samples, treatment with a regimen not covered by the insurance plan or not paid via insurance claim, or treatment with a regimen not recommended by the CDC. However, given that this is an insured population, it is reasonable to assume that actual treatment rates are substantially higher and probably match diagnosis rates— diagnosis rate is a good proxy for treatment rate. Thus, one major assumption of the current analyses was that those diagnosed after screening were treated. The increasing trend in chlamydia (and gonorrhea) diagnosis compared with the decreasing PID diagnosis rate in this cohort suggests that the increased screening helped to identify and treat relatively more chlamydia (and/or gonorrhea) cases, thereby reducing overall PID diagnosis rate. The analysis conducted in the present study is subject to the usual limitations associated with ecologic studies (and the Granger causality technique). Most important,
656
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
this type of study can show an association between increased screening and reduced PID rates, but it cannot establish a causal connection. Because the data were limited to 72 observations (6 years of monthly data), this analysis is exploratory in nature and should be revisited in the future as more data become available. Furthermore, the regression results presented in this study can be sensitive to the model specifıcation (e.g., the fourth versus the third/fıfth lag of the screening value; log versus nonlog values). This study did not specifıcally examine the potential indirect benefıt of chlamydia (and gonorrhea) screening as it relates to reduction in prevalence for the cohort. That is, this study focused on a cohort and ignored sexual mixing occurring between the cohort and the rest of the at-risk population. The use of diagnostic codes to estimate the incidence of PID can be problematic, given the lack of consensus on the exact symptoms, clinical signs, or laboratory result or any combination that is distinctively characteristic of PID cases.1 Previous studies22,23 have reported high percentages of falsely attributed PID cases when diagnostic codes are used to identify cases of PID meeting a clearly defıned case defınition. An earlier study22 reported a positive predictive value (PPV) of 18.1% for Diagnosis Code 614.9 using the CDC case defınition for PID. Code 614.9 made up more than half of the diagnoses in the current cohort (see Table 1), although its proportion decreased by approximately 15% during the period examined. However, to the extent that the PPV for the set of codes used was fairly constant over the time that the analyses covered, the current results would be insensitive to the PPV because this study focused on relative (rather than absolute) changes in PID diagnosis rates. There are at least two other potential chlamydia and gonorrhea testing codes (87801 and 87800) that were not included, primarily because they may also be used to detect organisms other than chlamydia and gonorrhea.24 –26 Also, electronic medical record systems in some jurisdictions have different sets of codes for chlamydia and gonorrhea screening. For instance, Massachusetts used 87800 and 87591 for chlamydia.25 Thus, the exclusion of some codes may have caused us to underestimate screening rates for chlamydia and gonorrhea. Finally, there might be other reasons for declines in PID in the current study cohort such as changes in diagnostic coding by providers. Although the diagnosis rate trends for chlamydia and gonorrhea are consistent with annual trends in surveillance report,27 the rates found in this study were substantially lower (⬎70%). Nonspecifıc coding may be one of the reasons for the relatively small number of cases found. Another major reason is that the insured population belongs to the group with higher SES, who have relatively lower STD rates.27,28
This study also has important strengths. The study used data from one of the largest administrative claims databases in the U.S., which has become an increasingly important part of recent advances in health services research.29 Second, the current data and analyses made it possible to estimate the joint effect of both chlamydia and gonorrhea screening on PID diagnosis. Given that both may cause PID and the recommendations to screen and treat for both were fırst issued as far back as 1985,30 it is important to assess their joint effect. This study applied a classic time–series technique that had not previously been used to examine the effect of chlamydia and gonorrhea screening on PID incidence. The predictive nature of the current results may be helpful in providing population-level information regarding potential PID outcomes from implementing screening programs or changes to existing programs. The age-specifıc analyses performed confırmed the robustness of the fınal results and provided evidence consistent with a causal, rather than coincidental, interpretation of the association observed in this study. If the changes in screening coverage rates and the changes in PID diagnosis rates were reflective of unrelated, population-level trends, one would have expected changes in screening coverage in the younger age group to be associated with changes in PID diagnosis in the older age group (and for changes in screening coverage in the older age group to be associated with changes in PID in the younger age group), but this was not the case. The signifıcance of the 4-month lag of both series may have some implications for chlamydia/gonorrhea screening and the epidemiology of chlamydia/gonorrhea-associated PID diagnosis. It represents a crude estimate of the time to expect the effects (i.e., changes in PID diagnosis rate) of changes to a chlamydia/gonorrhea screening program in a particular population. Second, knowledge about the period between chlamydial and/or gonococcal infections and the onset of symptomatic PID is scanty. Studies that could provide more reliable information in this area are considered unethical.10 The methods employed in this study could be used to obtain information about the natural history of PID, such as the time from chlamydia/gonorrhea infection to the onset of chlamydia/gonorrhea-associated PID. For example, the signifıcant association between increases in chlamydia/gonorrhea screening and decreases in PID diagnosis 4 months later might suggest that most infections require at least 4 months to progress to PID. A more-detailed analysis could examine this interpretation more rigorously. The fındings and conclusions in this manuscript are those of the authors and do not necessarily represent the views www.ajpm-online.net
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
of the CDC. Mention of company names or products does not imply endorsement by CDC. CKK received an honorarium for a presentation on HIV RNA testing in October 2005 from Gen-Probe, the largest manufacturer of chlamydia and gonorrhea testing. No other fınancial disclosures were reported by the authors of this paper.
References 1. Paanoveen J, Westrom L, Eschenbach DA. Pelvic inflammatory disease. In: Holmes KK, Sparling PF, Stamm WE, et al., ed. Sexually transmitted diseases. New York: McGraw Hill, 2008. 2. Cates W, Rolfs RT, Aral SO. Sexually transmitted diseases, pelvic inflammatory disease, and infertility—an epidemiologic update. Epidemiol Rev 1990;12:199 –220. 3. Scholes D, Stergachis A, Heidrich FE, Andrilla H, Holmes KK, Stamm WE. Prevention of pelvic inflammatory disease by screening for cervical chlamydial infection. N Engl J Med 1996;334(21):1362– 6. 4. Ostergaard L, Andersen B, Moller JK, Olesen F. Home sampling versus conventional swab sampling for screening of Chlamydia trachomatis in women: a cluster-randomized 1-year follow-up study. Clin Infect Dis 2000;31(4):951–7. 5. Bloom MS, Hu Z, Gaydos JC, Brundage JF, Tobler SK. Incidence rates of pelvic inflammatory disease diagnoses among Army and Navy recruits—potential impacts of chlamydia screening policies. Am J Prev Med 2008;34(6):471–7. 6. Low N. Screening programmes for chlamydial infection: when will we ever learn? BMJ 2007;334(7596):725– 8. 7. Morre SA, van den Brule AJC, Rozendaal L, et al. The natural course of asymptomatic Chlamydia trachomatis infections: 45% clearance and no development of clinical PID after one-year follow-up. Int J STD AIDS 2002;13(2S):12– 8. 8. Risser WL, Risser JMH. The incidence of pelvic inflammatory disease in untreated women infected with Chlamydia trachomatis: a structured review. Int J STD AIDS 2007;18(11):727–31. 9. Wallace LA, Scoular A, Hart G, Reid M, Wilson P, Goldberg DJ. What is the excess risk of infertility in women after genital chlamydia infection? A systematic review of the evidence. Sex Transm Infect 2008; 84(3):171–5. 10. Simms I, Hornert P. Has the incidence of pelvic inflammatory disease following chlamydial infection been overestimated? Int J STD AIDS 2008;19(4):285– 6. 11. Bohm MM, Rollins School of Public Health. The association between screening for cervical Chlamydia trachomatis infection and the diagnosis of pelvic inflammatory disease among privately insured 15 to 25 year old females in the U.S., 2001–2003 [Thesis (MPH)]: Emory University, 2006. 12. Hillis SD, Wasserheit JN. Screening for chlamydia—a key to the prevention of pelvic inflammatory disease. N Engl J Med 1996;334(21): 1399 – 401.
June 2010
657
13. Sims C. Money, income and causality. Am Econ Rev 1972;62(4): 540 –52. 14. Granger CWJ. Investigating causal relations by econometrics models and cross-spectral methods. Econometrica 1969;37(3):424 –38. 15. Dhamala M, Rangarajan G, Ding M. Analyzing information flow in brain networks with nonparametric Granger causality. Neuroimage 2008;41(2):354 – 62. 16. Elsner JB. Granger causality and Atlantic hurricanes. Tellus A Dyn Meteorol Oceanogr 2007;59(4):476 – 85. 17. Londei A, D’Ausilio A, Basso D, et al. Brain network for passive word listening as evaluated with ICA and Granger causality. Brain Res Bull 2007;72(4 – 6):284 –92. 18. Moineddin R, Nie JX, Domb G, Leong AM, Upshur REG. Seasonality of primary care utilization for respiratory diseases in Ontario: a timeseries analysis. BMC Health Serv Res 2008;8:160. 19. Wang QX, Liu Y, Pan XC. Atmosphere pollutants and mortality rate of respiratory diseases in Beijing. Sci Total Environ 2008;391(1):143– 8. 20. Zhou ZY, Chen YH, Ding MZ, Wright P, Lu ZH, Liu Y. Analyzing brain networks with PCA and conditional Granger causality. Hum Brain Mapp 2009;30(7):2197–206. 21. CDC, Workowski KA, Berman SM. Sexually transmitted disease treatment guidelines, 2006. MMWR Recomm Rep 2006;55(RR-11):1–94. 22. Ratelle S, Yokoe D, Blejan C, et al. Predictive value of clinical diagnostic codes for the CDC case defınition of pelvic inflammatory (PID)— implications for surveillance. Sex Transm Dis 2003;30(11):866 –70. 23. Simms I, Warburton F, Westrom L. Diagnosis of pelvic inflammatory disease: time for a rethink. Sex Transm Infect 2003;79(6):491– 4. 24. De Ravello L, Brantley MD, Lamarre M, Qayad MG, Aubert H, BeckSague C. Sexually transmitted infections and other health conditions of women entering prison in Georgia, 1998 –1999. Sex Transm Dis 2005;32(4):247–51. 25. Klompas M. Automated detection and reporting of notifıable diseases in near real-time using electronic medical records. 2007. www2.cdc. gov/ncphi/disss/seminars/archive.asp. 26. Owusu-Edusei K, Bohm MK, Kent CK. Diagnostic methodologies for chlamydia screening in females aged 15 to 25 years from private insurance claims data in the U.S., 2001 to 2005. Sex Transm Dis 2009; 36(7):419 –21. 27. CDC. Sexually Transmitted Disease Surveillance, 2007. Atlanta GA: USDHHS, 2008. 28. DeNavas-Walt C, Proctor DB, Smith JC, U.S. Census Bureau. Income, poverty, and health insurance coverage in the U.S.: 2007. In: Current Population Reports, P60-235. Washington DC: U.S. Government Printing Offıce, 2008. 29. Virnig BA, McBean M. Administrative data for public health surveillance and planning. Annu Rev Public Health 2001;22:213–30. 30. CDC. Sexually transmitted disease treatment guidelines. MMWR Recomm Rep 1985;34:75S–108S.
Appendix Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.amepre. 2010.02.008.
Purpose: This study aims to examine the association between chlamydia and gonorrhea screening and PID diagnoses using time–series analyses. Methods: Monthly chlamydia and gonorrhea screening and PID diagnosis rates were extracted for a cohort of 207,695 continuously enrolled privately insured women from January 2001 to December 2006. An autoregressive integrated moving average model was used to examine whether rates of PID diagnoses in a given month were associated with rates of chlamydia and gonorrhea screening in previous months.
Results: Monthly screening rates increased from about 300 to almost 700 per 100,000 for chlamydia and from 250 to almost 650 per 100,000 for gonorrhea, whereas PID diagnosis rates declined during the same period (40 –20 per 100,000). Increases in screening rates were associated with decreases in PID diagnosis rates 4 months later. On average, a one-unit (or 10%) increase in the growth of chlamydia and gonorrhea screening rates, separately, in the prior fourth month was signifıcantly associated with a 0.36 (or 3.6%, p⬍0.05) and 0.32 (or 3.2%, p⬍0.10) decrease in the growth rate of the PID diagnosis rate, respectively. Conclusions: Although analyses such as these cannot prove causality, the results are consistent with the hypothesis that increases in chlamydia and gonorrhea screening coverage can lead to reductions in PID at the population level. A population-level focus offers advantages over individual-level analyses of screening and PID, such as the ability to capture indirect benefıts of increased screening. (Am J Prev Med 2010;38(6):652– 657) Published by Elsevier Inc. on behalf of American Journal of Preventive Medicine
Introduction
P
elvic inflammatory disease (PID) has been broadly described1 as a range of upper genital tract disorders among women, including one or a combination of the following: endometritis, salpingitis, tubo-ovarian abscess, and pelvic peritonitis. PID is caused by the introduction of pathogenic micro-organisms from the lower to the upper reproductive organs.2 It develops in 10% to 40% of untreated chlamydial or gonococcal cervicitis cases in women.1 Thus, although gonorrhea (GC) and chlamydia are known common causes of PID, a substan-
From the Division of STD Prevention, CDC, Atlanta, Georgia Address correspondence to: Kwame Owusu-Edusei Jr, PhD, Division of STD Prevention, CDC, 1600 Clifton Road, MS E-80, Atlanta GA 30333. E-mail: [email protected]. 0749-3797/00/$17.00 doi: 10.1016/j.amepre.2010.02.008
652
Am J Prev Med 2010;38(6):652– 657
tial proportion of PID cases have other causes or no known etiology.1 A seminal RCT published in 1996 found that selective testing for chlamydia in at-risk women reduced the incidence of PID.3 Since then, two other studies4,5 have reported comparable results in different settings using different methods. However, potential limitations of these studies have been noted and the magnitude of the utility of chlamydia screening as a tool to prevent PID has subsequently been debated.6 –10 Past analyses11 of claims data have found that young women diagnosed with PID were more likely to have been previously screened for chlamydia than young women without a PID diagnosis. This positive association between chlamydia screening and PID (which seems at odds with the idea that chlamydia screening can reduce PID rates12) likely arises because (1) screening is targeted
Published by Elsevier Inc. on behalf of American Journal of Preventive Medicine
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
toward those with risk factors for chlamydia and (2) a notable portion of the screening tests might actually be diagnostic tests among those seeking care as a result of clinical symptoms.11 Because these important individuallevel confounders that influence the risk of PID cannot be controlled for in analyses using claims data, this study focused on changes in screening coverage rates and changes in PID diagnosis rates at the population level using time–series analyses. A population-level approach reduces the confounding of the association between screening and PID at the individual level and allows us to address the question of whether changes in the frequency of screening can affect PID rates at the population level. In addition, a population-level focus has the advantage of being able to capture indirect benefıts (such as interruption of disease transmission and reduced population prevalence) of chlamydia and gonorrhea screening and treatment on PID rates. One method used to investigate causality (or association) for time–series data is the Granger causality technique.13 According to Granger,14 if past measures of a variable/event (X) helps to predict another (Y), then X “Granger causes” Y. Following its application to economic data in 1972,13 several other fıelds, including medicine, have employed it to study relationships between data sets over time.15–20 The authors are unaware of the application of the Granger causality technique to study
653
sexually transmitted diseases and their sequelae. Application of this technique (i.e., Granger-causality) may potentially assist and/or complement results/analyses from public health surveillance data and other data sources such as claims data (particularly as the nation moves to richer and more extensive electronic medical record systems). In this study, the Granger causality technique was used to investigate potential population-level associations between monthly chlamydia/gonorrhea screening and PID diagnosis.
Methods
Outpatient and inpatient claims data from the MarketScan Database (MEDSTAT Group) for 2001 through 2006 were used. The database contains claims data on more than 5 million people who have employer-sponsored health insurance from more than 100 payers, including large employers, health plans, and government and public organizations. The employer-sponsored privately insured constitute about 58% of the population in the U.S. and analyses from the MarketScan Database can be projected to represent that population (Thompson Medstat Group). Monthly claims data were extracted for a cohort of 207,695 women (aged 15–39 years in 2001) who were enrolled continuously from January 1, 2001, through December 31, 2006 —72 months. Each of the women in the cohort was enrolled for a total of 2191 member-days (i.e., 365 days for each year except 2004, which was a leap year). Current procedural terminology (CPT) codes for chlamydia screening (87110, 87270, 87320, 87490, 87491, and 87810) and gonorrhea screening (87590, 87591, 87592, and 87850) were used to identify the number of women in the cohort who were screened Table 1. Number and distribution of PID diagnosis codes each month. The ICD-9 codes for chlamydia Code Description Frequencya Percentage (078.88, 079.88, 079.98, 099.41, and 099.50 – 098.10 Acute GC upper GU tract, site unspecified 7 0.04 099.59) and gonorrhea (098.0 – 098.89) diagnoses 098.16 Acute GC endometritis 15 0.09 were used to construct 098.17 Acute GC salpingitis 39 0.24 and examine monthly di098.19 Acute GC upper GU tract, other site 30 0.19 agnosis rates. National Drug Codes (NDC) of 098.86 Acute GC peritonitis 64 0.40 drugs recommended by 099.56 Acute chlamydia peritonitis 2 0.01 the CDC for treating chlamydia and gonorrhea21 614.0 Acute salpingo-oophoritis 656 4.10 from drug claims tables 614.2 Salpingitis/oophoritis, not acute or chronic 1950 12.20 were used to estimate treatment rates. 614.3 Acute parametritis/PID 1680 10.51 Women in the cohort 614.5 Acute or unspecified pelvic peritonitis 10 0.06 diagnosed with acute PID in each month were 614.8 Other specified inflammatory disease, female pelvic organs 119 0.74 identifıed using ICD-9 614.9 Unspecified inflammatory disease, female pelvic organs 8205 51.31 codes (Table 1) from 615.0 Inflammatory disease of uterus, except cervix 787 4.92 the inpatient and outpatient claims tables. 615.9 Unspecified inflammatory disease of uterus 2426 15.17 The time–series analyTotal 15,990 100 ses performed in this a study examined whether Note that in a substantial number of instances, more than one diagnosis code was associated with an changes in chlamydia and enrollee in 1 month. gonorrhea screening rates GC, gonorrhea; GU, genitourinary; PID, pelvic inflammatory disease
June 2010
654
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
Figure 1. Monthly chlamydia and gonorrhea screening rates (per 100,000) and pelvic inflammatory disease (PID) diagnosis rates (per 100,000) for a cohort of employersponsored privately insured women aged 15–39 years in 2001 who were continuously enrolled from January 2001 through December 2006. Note that the vertical axes have different scales.
creased substantially (from their lowest points in the early months of 2001) to the end of the period examined; chlamydia screening rates increased from about 300 to almost 700 (a 133% increase) and gonorrhea screening rates increased from about 250 to almost 650 (a 160% increase). The monthly PID diagnosis rate decreased from about 40 to 20 (a 50% decrease) during the same period. The screening rate for chlamydia was 4%–28% higher than the screening rate for gonorrhea in each month. Thus, for this cohort, chlamydia and gonorrhea screening rates trended upward, whereas the PID diagnosis rates trended downward (see Figure 1). In addition, the chlamydia and gonorrhea series had similar trends, with a signifıcant (p⬍0.001) high contemporaneous correlation coeffıcient of 0.97. Diagnosis rates for both chlamydia and gonorrhea increased for each subsequent year. For the cohort examined, the diagnosis rate (number per 100,000) for chlamydia increased from about 5 in January 2001 to about 13 in December 2006, and from 2 in January 2001 to about 6 in December 2006 for gonorrhea. However, the treatment rate for the cohort was notably low (⬍40%). Results of the regression from models 1, 2, and 3 are presented in Table 2. More detailed information on the time–series regression results has been provided in Appendix B (available online at www.ajpm-onlinenet). In both cases, the fourth lags (4-month lag between screening and PID diagnosis) were signifıcant (p⬍0.10), although the coeffıcient for chlamydia was signifıcant at 5% (see Table 2). The coeffıcients for the fourth lag of chlamydia indicated that a one-unit (or 10%) increase in the growth rate of chlamydia screening rate in the past fourth month was associated with a 0.36 (or 3.6%) decrease in the growth rate of PID diagnosis, on average. Similarly, a one-unit (or
in a given month were associated with changes in PID rates in subsequent months. See Appendix A (available online at www.ajpmonline.net) for details on the time–series analyses. The signifıcance level was set at 10% because of the relatively small number of observations (n⫽72 months). The analysis was repeated for two mutually exclusive age categories—women aged 15–24 years and 31–39 years in 2001. As a test against spurious correlations, examination of whether changes in screening coverage rates in the younger cohort were associated with changes in PID diagnosis rates in the older cohort, and whether changes in screening coverage rates in the older cohort were associated with changes in PID rates in the younger cohort, were conducted. The idea behind this additional test was that if screening is causally associated with subsequent reductions in PID diagnosis, then changes in screening coverage in the younger (or older) cohort would be expected to have much more of an impact on PID rates in that cohort than in the Table 2. Factors associated with monthly PID diagnosis rates: Time–series regression other cohort. The analyanalysis coefficients (t statistics) ses were conducted in 2008 and 2009 using Variable Model 1 Model 2 Model 3 SAS, version 9.2, and Eviews, version 6. Constant ⫺0.02 (⫺1.84)* ⫺0.01 (⫺1.99)* ⫺0.01 (⫺1.92)*
Results Figure 1 shows monthly screening rates (number screened per 100,000) for chlamydia and gonorrhea, and PID diagnosis rates (number of cases per 100,000). Chlamydia and gonorrhea screening rates in-
PID rate, month t-1
⫺0.92 (⫺7.8)***
⫺0.98 (⫺12.9)***
⫺0.99 (⫺12.6)***
PID rate, month t-2
⫺0.53 (⫺4.5)***
⫺0.52 (⫺8.6)***
⫺0.52 (⫺8.0)***
Chlamydia screening rate, month t-4
—
Gonorrhea screening rate, month t-4
—
Adjusted R2
0.53
⫺0.36 (⫺2.0)** — 0.65
— ⫺0.32 (⫺1.9)* 0.65
Note: Regression coefficients and t statistics (in parentheses) are shown. PID rates and screening rates for chlamydia and gonorrhea were expressed in terms of the change in the log value from the previous month. A complete version of this table is provided in Appendix B (available online at www.ajpm-online.net). *0.1, **0.05, ***0.01 PID, pelvic inflammatory disease
www.ajpm-online.net
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
655
0.28 (p⬍0.10) decrease in the growth rate of PID diagnosis for the younger and older age groups, respectively (results not shown). In testing the model for susceptibility to spurious associations, this study found that changes in chlamydia screening rates in the younger age cohort were not associated with changes in PID rates in the older age cohort, and changes in chlamydia screening rates in the older age group were not associated with changes in PID in the younger age group (p⬎0.3).
Discussion Figure 2. Monthly chlamydia screening rate (per 100,000) and pelvic inflammatory disease (PID) diagnosis rates (number per 100,000) for employer-sponsored privately insured women aged 15–24 years and 31–39 years in 2001 who were continuously enrolled from January 2001 through December 2006. Note that the vertical axes have different scales.
10%) increase in gonorrhea screening rate in the past fourth month was associated with a 0.32 (or 3.2%) decrease in the growth rate of PID diagnosis, on average. As an example, if the growth rate of chlamydia screening increased by one unit, say from 6% to 7% (representing an increase from 400/100,000 to 428/100,000, instead of to 424/100,000), one would expect the growth rate of PID diagnosis 4 months later to decrease by 0.36 from ⫺3% to ⫺3.36% (representing a decrease from 30/100,000 to 28.92/ 100,000, instead of to 29.1/100,000). Similarly, a one-unit increase (6% to 7%) in the growth rate of gonorrhea screening would be expected to decrease the growth rate of PID diagnosis 4 months later by 0.32 from ⫺3% to ⫺3.32% (representing a decrease from 30/100,000 to 29.04/100,000, instead of to 29.1/100,000). Monthly screening rates for chlamydia and PID diagnosis rates for the two age groups (women aged 15–24 years and 31–39 years in 2001) are shown in Figure 2. The screening rate for the older age group showed a slight upward trend but was substantially lower than the screening rate for the younger age group. The PID diagnosis trend for the older age group was also substantially lower than for younger women and was fairly stable, vacillating at approximately 15 cases per 100,000 women. Chlamydia screening and PID diagnosis rates for younger women were substantially higher and exhibit noticeably steeper trends and higher variance. As with the base case analysis, changes in chlamydia screening rates helped to predict changes in the subsequent PID diagnosis rates within each age group. On average, a one-unit increase in the growth rate of chlamydia screening was associated with a 0.38 (p⬍0.10) and June 2010
To the authors’ knowledge, this is the fırst application of time–series analysis of medical claims data to explore the potential impact of chlamydia and gonorrhea screening in terms of preventing PID. This study found that increases in the 4-month lag of both the chlamydia and gonorrhea screening series improved the predictive power of the PID model. That is, screening rates in a given month helped to predict PID rates 4 months later. Thus, the results suggest that there are population-level benefıts from increases in chlamydia/gonorrhea screening in terms of reducing PID. Although this observational study design cannot establish a causal association between increased screening and decreased PID, the results are consistent with a causal association as well as with previously published studies3–5 showing that increased chlamydia/ gonorrhea screening was associated with reductions in the burden of PID. The thesis that chlamydia screening may reduce PID is based on the idea that the cases identifıed would be treated, subsequently preventing PID. This study found an unrealistically low treatment rate for those diagnosed with chlamydia and/or gonorrhea after screening. The low treatment rate may largely be due to one (or a combination) of these reasons: treatment with free samples, treatment with a regimen not covered by the insurance plan or not paid via insurance claim, or treatment with a regimen not recommended by the CDC. However, given that this is an insured population, it is reasonable to assume that actual treatment rates are substantially higher and probably match diagnosis rates— diagnosis rate is a good proxy for treatment rate. Thus, one major assumption of the current analyses was that those diagnosed after screening were treated. The increasing trend in chlamydia (and gonorrhea) diagnosis compared with the decreasing PID diagnosis rate in this cohort suggests that the increased screening helped to identify and treat relatively more chlamydia (and/or gonorrhea) cases, thereby reducing overall PID diagnosis rate. The analysis conducted in the present study is subject to the usual limitations associated with ecologic studies (and the Granger causality technique). Most important,
656
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
this type of study can show an association between increased screening and reduced PID rates, but it cannot establish a causal connection. Because the data were limited to 72 observations (6 years of monthly data), this analysis is exploratory in nature and should be revisited in the future as more data become available. Furthermore, the regression results presented in this study can be sensitive to the model specifıcation (e.g., the fourth versus the third/fıfth lag of the screening value; log versus nonlog values). This study did not specifıcally examine the potential indirect benefıt of chlamydia (and gonorrhea) screening as it relates to reduction in prevalence for the cohort. That is, this study focused on a cohort and ignored sexual mixing occurring between the cohort and the rest of the at-risk population. The use of diagnostic codes to estimate the incidence of PID can be problematic, given the lack of consensus on the exact symptoms, clinical signs, or laboratory result or any combination that is distinctively characteristic of PID cases.1 Previous studies22,23 have reported high percentages of falsely attributed PID cases when diagnostic codes are used to identify cases of PID meeting a clearly defıned case defınition. An earlier study22 reported a positive predictive value (PPV) of 18.1% for Diagnosis Code 614.9 using the CDC case defınition for PID. Code 614.9 made up more than half of the diagnoses in the current cohort (see Table 1), although its proportion decreased by approximately 15% during the period examined. However, to the extent that the PPV for the set of codes used was fairly constant over the time that the analyses covered, the current results would be insensitive to the PPV because this study focused on relative (rather than absolute) changes in PID diagnosis rates. There are at least two other potential chlamydia and gonorrhea testing codes (87801 and 87800) that were not included, primarily because they may also be used to detect organisms other than chlamydia and gonorrhea.24 –26 Also, electronic medical record systems in some jurisdictions have different sets of codes for chlamydia and gonorrhea screening. For instance, Massachusetts used 87800 and 87591 for chlamydia.25 Thus, the exclusion of some codes may have caused us to underestimate screening rates for chlamydia and gonorrhea. Finally, there might be other reasons for declines in PID in the current study cohort such as changes in diagnostic coding by providers. Although the diagnosis rate trends for chlamydia and gonorrhea are consistent with annual trends in surveillance report,27 the rates found in this study were substantially lower (⬎70%). Nonspecifıc coding may be one of the reasons for the relatively small number of cases found. Another major reason is that the insured population belongs to the group with higher SES, who have relatively lower STD rates.27,28
This study also has important strengths. The study used data from one of the largest administrative claims databases in the U.S., which has become an increasingly important part of recent advances in health services research.29 Second, the current data and analyses made it possible to estimate the joint effect of both chlamydia and gonorrhea screening on PID diagnosis. Given that both may cause PID and the recommendations to screen and treat for both were fırst issued as far back as 1985,30 it is important to assess their joint effect. This study applied a classic time–series technique that had not previously been used to examine the effect of chlamydia and gonorrhea screening on PID incidence. The predictive nature of the current results may be helpful in providing population-level information regarding potential PID outcomes from implementing screening programs or changes to existing programs. The age-specifıc analyses performed confırmed the robustness of the fınal results and provided evidence consistent with a causal, rather than coincidental, interpretation of the association observed in this study. If the changes in screening coverage rates and the changes in PID diagnosis rates were reflective of unrelated, population-level trends, one would have expected changes in screening coverage in the younger age group to be associated with changes in PID diagnosis in the older age group (and for changes in screening coverage in the older age group to be associated with changes in PID in the younger age group), but this was not the case. The signifıcance of the 4-month lag of both series may have some implications for chlamydia/gonorrhea screening and the epidemiology of chlamydia/gonorrhea-associated PID diagnosis. It represents a crude estimate of the time to expect the effects (i.e., changes in PID diagnosis rate) of changes to a chlamydia/gonorrhea screening program in a particular population. Second, knowledge about the period between chlamydial and/or gonococcal infections and the onset of symptomatic PID is scanty. Studies that could provide more reliable information in this area are considered unethical.10 The methods employed in this study could be used to obtain information about the natural history of PID, such as the time from chlamydia/gonorrhea infection to the onset of chlamydia/gonorrhea-associated PID. For example, the signifıcant association between increases in chlamydia/gonorrhea screening and decreases in PID diagnosis 4 months later might suggest that most infections require at least 4 months to progress to PID. A more-detailed analysis could examine this interpretation more rigorously. The fındings and conclusions in this manuscript are those of the authors and do not necessarily represent the views www.ajpm-online.net
Owusu-Edusei et al / Am J Prev Med 2010;38(6):652– 657
of the CDC. Mention of company names or products does not imply endorsement by CDC. CKK received an honorarium for a presentation on HIV RNA testing in October 2005 from Gen-Probe, the largest manufacturer of chlamydia and gonorrhea testing. No other fınancial disclosures were reported by the authors of this paper.
References 1. Paanoveen J, Westrom L, Eschenbach DA. Pelvic inflammatory disease. In: Holmes KK, Sparling PF, Stamm WE, et al., ed. Sexually transmitted diseases. New York: McGraw Hill, 2008. 2. Cates W, Rolfs RT, Aral SO. Sexually transmitted diseases, pelvic inflammatory disease, and infertility—an epidemiologic update. Epidemiol Rev 1990;12:199 –220. 3. Scholes D, Stergachis A, Heidrich FE, Andrilla H, Holmes KK, Stamm WE. Prevention of pelvic inflammatory disease by screening for cervical chlamydial infection. N Engl J Med 1996;334(21):1362– 6. 4. Ostergaard L, Andersen B, Moller JK, Olesen F. Home sampling versus conventional swab sampling for screening of Chlamydia trachomatis in women: a cluster-randomized 1-year follow-up study. Clin Infect Dis 2000;31(4):951–7. 5. Bloom MS, Hu Z, Gaydos JC, Brundage JF, Tobler SK. Incidence rates of pelvic inflammatory disease diagnoses among Army and Navy recruits—potential impacts of chlamydia screening policies. Am J Prev Med 2008;34(6):471–7. 6. Low N. Screening programmes for chlamydial infection: when will we ever learn? BMJ 2007;334(7596):725– 8. 7. Morre SA, van den Brule AJC, Rozendaal L, et al. The natural course of asymptomatic Chlamydia trachomatis infections: 45% clearance and no development of clinical PID after one-year follow-up. Int J STD AIDS 2002;13(2S):12– 8. 8. Risser WL, Risser JMH. The incidence of pelvic inflammatory disease in untreated women infected with Chlamydia trachomatis: a structured review. Int J STD AIDS 2007;18(11):727–31. 9. Wallace LA, Scoular A, Hart G, Reid M, Wilson P, Goldberg DJ. What is the excess risk of infertility in women after genital chlamydia infection? A systematic review of the evidence. Sex Transm Infect 2008; 84(3):171–5. 10. Simms I, Hornert P. Has the incidence of pelvic inflammatory disease following chlamydial infection been overestimated? Int J STD AIDS 2008;19(4):285– 6. 11. Bohm MM, Rollins School of Public Health. The association between screening for cervical Chlamydia trachomatis infection and the diagnosis of pelvic inflammatory disease among privately insured 15 to 25 year old females in the U.S., 2001–2003 [Thesis (MPH)]: Emory University, 2006. 12. Hillis SD, Wasserheit JN. Screening for chlamydia—a key to the prevention of pelvic inflammatory disease. N Engl J Med 1996;334(21): 1399 – 401.
June 2010
657
13. Sims C. Money, income and causality. Am Econ Rev 1972;62(4): 540 –52. 14. Granger CWJ. Investigating causal relations by econometrics models and cross-spectral methods. Econometrica 1969;37(3):424 –38. 15. Dhamala M, Rangarajan G, Ding M. Analyzing information flow in brain networks with nonparametric Granger causality. Neuroimage 2008;41(2):354 – 62. 16. Elsner JB. Granger causality and Atlantic hurricanes. Tellus A Dyn Meteorol Oceanogr 2007;59(4):476 – 85. 17. Londei A, D’Ausilio A, Basso D, et al. Brain network for passive word listening as evaluated with ICA and Granger causality. Brain Res Bull 2007;72(4 – 6):284 –92. 18. Moineddin R, Nie JX, Domb G, Leong AM, Upshur REG. Seasonality of primary care utilization for respiratory diseases in Ontario: a timeseries analysis. BMC Health Serv Res 2008;8:160. 19. Wang QX, Liu Y, Pan XC. Atmosphere pollutants and mortality rate of respiratory diseases in Beijing. Sci Total Environ 2008;391(1):143– 8. 20. Zhou ZY, Chen YH, Ding MZ, Wright P, Lu ZH, Liu Y. Analyzing brain networks with PCA and conditional Granger causality. Hum Brain Mapp 2009;30(7):2197–206. 21. CDC, Workowski KA, Berman SM. Sexually transmitted disease treatment guidelines, 2006. MMWR Recomm Rep 2006;55(RR-11):1–94. 22. Ratelle S, Yokoe D, Blejan C, et al. Predictive value of clinical diagnostic codes for the CDC case defınition of pelvic inflammatory (PID)— implications for surveillance. Sex Transm Dis 2003;30(11):866 –70. 23. Simms I, Warburton F, Westrom L. Diagnosis of pelvic inflammatory disease: time for a rethink. Sex Transm Infect 2003;79(6):491– 4. 24. De Ravello L, Brantley MD, Lamarre M, Qayad MG, Aubert H, BeckSague C. Sexually transmitted infections and other health conditions of women entering prison in Georgia, 1998 –1999. Sex Transm Dis 2005;32(4):247–51. 25. Klompas M. Automated detection and reporting of notifıable diseases in near real-time using electronic medical records. 2007. www2.cdc. gov/ncphi/disss/seminars/archive.asp. 26. Owusu-Edusei K, Bohm MK, Kent CK. Diagnostic methodologies for chlamydia screening in females aged 15 to 25 years from private insurance claims data in the U.S., 2001 to 2005. Sex Transm Dis 2009; 36(7):419 –21. 27. CDC. Sexually Transmitted Disease Surveillance, 2007. Atlanta GA: USDHHS, 2008. 28. DeNavas-Walt C, Proctor DB, Smith JC, U.S. Census Bureau. Income, poverty, and health insurance coverage in the U.S.: 2007. In: Current Population Reports, P60-235. Washington DC: U.S. Government Printing Offıce, 2008. 29. Virnig BA, McBean M. Administrative data for public health surveillance and planning. Annu Rev Public Health 2001;22:213–30. 30. CDC. Sexually transmitted disease treatment guidelines. MMWR Recomm Rep 1985;34:75S–108S.
Appendix Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.amepre. 2010.02.008.