Share this article on:

Prediction of Pelvic Inflammatory Disease Among Young, Single, Sexually Active Women

Ness, Roberta B. MD, MPH*; Smith, Kenneth J. MD; Chang, Chung-Chou H. PhD; Schisterman, Enrique F. PhD; Bass, Debra C. MS*for the Gynecologic Infection Follow-Through (GIFT) Investigators

Sexually Transmitted Diseases: March 2006 - Volume 33 - Issue 3 - p 137-142
doi: 10.1097/01.olq.0000187205.67390.d1

Objectives: To assess prediction strategies for pelvic inflammatory disease (PID).

Study Design: One thousand one hundred seventy women were enrolled based on a high chlamydial risk score. Incident PID over a median of 3 years was diagnosed by either histologic endometritis or Centers for Disease Control and Prevention criteria. A multivariable prediction model for PID was assessed.

Results: Women enrolled using the risk score were young, single, sexually active, and often had prior sexually transmitted infections. Incident PID was common (8.6%). From 24 potential predictors, significant factors included age at first sex, gonococcal/chlamydial cervicitis, history of PID, family income, smoking, medroxyprogesterone acetate use, and sex with menses. The model correctly predicted 74% of incident PID; in validation models, correct prediction was only 69%.

Conclusions: Our data validate a modified chlamydial risk factor scoring system for prediction of PID. Additional multivariable modeling contributed little to prediction. Women identified by a threshold value on the chlamydial risk score should undergo intensive education and screening.

A modified chlamydial scoring system identified women at high risk for developing pelvic inflammatory disease (PID). Multivariable models added little additional prediction, validating that simple identification of chlamydial risk best predicts PID.

From the *Graduate School of Public Health, University of Pittsburgh, Pittsburgh, Pennsylvania; †Division of General Internal Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania; and the ‡National Institutes of Child Health and Disease, Bethesda, Maryland

We gratefully acknowledge the funding source: AI44151-01 from the National Institutes of Allergy and Infectious Disease and the GIFT investigators: Sharon L. Hillier, PhD; James A. McGregor, MD; Peter Rice, MD; Holly E. Richter, PhD, MD; David E. Soper, MD; Carol A. Stamm, MD; Richard L. Sweet, MD. We thank the following individuals whose dedication to working with the women enrolled in GIFT made this study possible: Susie Alagasarmy, Julie Beuler, Debbie Carr, Hope Cohen-Webb, Leslie Curll, Christine Donahue, Amanda Farmer, RN, Janice French, Melissa Girman, Alice Howell, RN, Juliette Hunt, Ellen Klein, BS, Faye LeBoeuf, CNM, April Lehman, Rosalyn Liu, Ingrid Macio, PA, Kathleen McKenna, Kim Miller, RN, Megan Mundy, Anne Rideout, CRNP, Jacqueline Travasso, Jennifer Watts, Casey Zuckerman, Katherine Simpson, and Barbara Kolodjiez.

Correspondence: Roberta B. Ness, University of Pittsburgh, Graduate School of Public Health, Room A548 Crabtree Hall, 130 DeSoto Street, Pittsburgh, PA 15261. E-mail:

Received for publication June 8, 2005, and accepted August 9, 2005.

CLINICAL PREDICTION MODELS ARE an efficient way of identifying individuals in whom to concentrate prevention efforts. Multivariable approaches have successfully characterized individuals at risk for developing a variety of conditions, including heart disease, breast cancer, and lung cancer.1–3

Pelvic inflammatory disease (PID) is a disease for which multivariable models have been developed to improve the diagnosis among women with acute signs and symptoms.4,5 These models have not successfully identified an algorithm with high diagnostic accuracy, although they have shown that some clinical features enhance sensitivity, whereas others improve specificity. Somewhat better risk stratification has been attained for detecting concurrent chlamydial infection. Stergachis et al.6 proposed a risk factor scoring system which showed sensitivities of 75% to 88% and specificities of 57% to 74%, depending on the model parameters used. Subsequent use of this model in a randomized screening trial by Scholes et al.7 identified high-risk women whose 1-year rate of PID was over 2%.

Few studies have attempted to predict future risk for developing PID. Individual, independent risk factors have been identified in retrospective and cross-sectional studies.8–14 These have included, most consistently, young age, low socioeconomic status, cigarette smoking, delayed heath care seeking behavior, concurrent Chlamydia trachomatis infection, and possibly lack of barrier and hormonal contraception, race, parity, and douching. To our knowledge, only 2 studies, both among sex workers in Nairobi, Kenya, examined risk factors in prospectively collected data.15,16 One involved a comparison of women with chlamydial PID versus women with chlamydial cervicitis, an approach that characterized factors related to ascension of chlamydia, rather than risk of future PID.15 The other focused solely on contraceptive practices.16

The current investigation examines the usefulness of the Stergachis et al.6 chlamydia-risk scoring system for prediction of PID. It further attempts to combine a variety of risk factors and markers into a single, parsimonious model for better predicting PID. We used data from a cohort study of young, single, sexually active women who met an at-risk cutoff on a modified Stergachis risk score. These women were followed to the endpoint of PID or until the end of the study period.17 Using Cox proportional hazards survival models to formulate the clinical prediction rule and multiple resampling of the data to validate the rule, we attempted to develop a model that would accurately predict future risk for developing PID.

Back to Top | Article Outline


Patient Selection

Women 13 to 36 years of age were recruited consecutively into the Gynecologic Infections Follow-through (GIFT) Study from family planning clinics, university health clinics, gynecology clinics, and STD units at each of 5 clinical sites located throughout the eastern, southern, and western regions of the United States between May 1999 and June 2001.18 Human subjects approval was obtained at each participating institution, and all women signed informed consent. Women were eligible for the GIFT study if they were not specifically seeking care for an STD but they were at elevated risk for having chlamydial cervicitis, according to a modification of the Stergachis et al.6 risk paradigm. Specifically, to be enrolled, a woman had to have a score of 3 points or more on an algorithm wherein points were derived as follows: age 24 or less = 1; black race = 2; never pregnant = 1; 2 or more sexual partners = 1; douches at least once per month = 2; any prior sexually transmitted infection, including N gonorrhoeae, C trachomatis, and T vaginalis = 2. For example, eligibility could be met by being young and black; young with a prior sexually transmitted infection; nonmonogomous and douching, etc. Of 2740 women screened for study entry, 853 (31.1%) did not meet these inclusion criteria. An additional 259 (9.5%) women were excluded on the basis of a priori criteria, including currently pregnant by β hCG testing; currently married; never having had sexual intercourse; having pelvic tenderness on examination at baseline; having had a hysterectomy, salpingectomy, or tubal ligation; or being on antibiotics at baseline. Among the 1628 women who were eligible for the study, 1199 (73.6%) completed a baseline questionnaire.

Back to Top | Article Outline


Among the 1199 study subjects, 29 (2.4%) had a baseline visit only. The remaining 1170 subjects are the focus of these analyses. They had a median length of follow-up of 3.0 years (interquartile range: 2.4 years to 3.4 years), and 88% completed their final scheduled interview.

Back to Top | Article Outline

Reported Risk Characteristics at Baseline

In a standardized 20-minute interview conducted by trained research staff at each center at baseline and then every 6 months, women were interviewed about demographic, lifestyle, and medical information. Women were also asked about demographic factors, including age, race, education, income, and gravidity. They also reported relevant lifestyle behaviors such as tobacco smoking, number of sexual partners in the past 2 months, acquisition of a new partner in the past 2 months, contraceptive use, sex with menses, and douching and douching frequency. Furthermore, they were requested to recall past episodes of sexually transmitted infections, including PID and gonococcal and/or chlamydial genital infections. To elicit a history of PID, women were asked, “Has a doctor or nurse ever told you that you had a pelvic infection, pelvic inflammatory disease, or PID? This is also known as an infection of the tubes, ovaries, or womb or ‘pus tubes'.”

Back to Top | Article Outline

Gonococcal/Chlamydial Genital Infections at Baseline

Educated by study staff at baseline on a standardized method for self-collection of vaginal specimens using a Q-tip-like cotton swab, subjects collected baseline specimens.19 DNA amplification for N gonorrhoeae and C trachomatis was performed using a strand displacement DNA Amplification (SDA) Assay (Becton Dickinson, Sparks, MD) from self-obtained vaginal swabs. In the first 450 women, SDA testing of urine was also accomplished and was found to correlate with vaginal swab SDA 100% of the time. All positive test results for gonococcal or chlamydial infection were reported to the clinical sites within 1 week of testing, where appropriate antibiotics were prescribed.

Back to Top | Article Outline


To detect PID, women who experienced pelvic pain at any point in the study and women who tested positive on N gonorrhoeae or C trachomatis self-collected screening, repeated at 6- to 12-month intervals, were scheduled for an additional, symptomatic visit involving a pelvic examination and an endometrial biopsy. PID was categorized upon finding (1) endometritis, a histologic diagnosis based on a modification20 of the criteria proposed by Kiviat et al.21 involving identification of at least 5 neutrophils in the endometrial surface epithelium, in the absence of menstrual endometrium and/or at least 2 plasma cells in the endometrial stroma on a hematoxylin and eosin-stained and methyl green pyronine-stained endometrial tissue slide; or (2) the presence of the following: a complaint of pelvic discomfort of less than 4 weeks' duration; a pelvic tenderness score, using the McCormack scale,22 of 1 or more; and the presence of oral temperature >101°F (38.3°C), leukorrhea or mucopus, erythrocyte sedimentation rate >15 mm/hour, white blood cell count >10,000, or gonococcal or chlamydial cervicitis.23 One hundred one women developed PID, half of whom had endometritis.

Back to Top | Article Outline

Statistical Analysis

Means and standard deviations were calculated for participants' characteristics by PID outcome status. We counted as a positive outcome only the first episode of incident PID experienced by a given woman. To evaluate the statistical differences between the disease status categories, we implemented separately for each comparison a univariable Cox proportional hazards model, after using the method of Grambsch and Therneau24 to test the validity of the proportional hazards assumption. Cox proportional hazards modeling is a type of survival analysis and was chosen because it allows the appropriate handling of censored observations.

Parameter estimation was performed on the entire data set containing all variables whose univariable proportional hazards model likelihood ratio test P values were <0.2 with stepwise removal from the model when P >0.10. These lenient entry and exit criteria enhanced the number of variables that entered and remained in the model. The likelihood ratio, which was used to determine the predictiveness of variables, represents the proportion of women who develop PID versus the proportion who do not develop PID among those with a positive test.25 That is, the objective was to maximize true positives and minimize false positives predicted by our test. Goodness of fit of the final model was tested using Cox-Snell residual plots. The c-statistic, which corresponds directly to the area under the receiver operating curve (ROC) for the model, was calculated for the final model and its 95% confidence interval obtained by bootstrapping (repeating the modeling) for 1000 repetitions.

Data were examined in both the format in which they were collected and in dichotomized format, with dichotomization performed based on maximizing the likelihood ratio in univariable Cox models. Cross-validation of the model to determine its robustness was performed using a grouped jackknife approach. Using a random-number generator, all patient data were divided into 10 independent subgroups, each subgroup containing 10% of the data. Nine of the 10 subgroups were used to fit a Cox model using the stepwise procedure above, and then this model used to predict outcomes in the excluded data. This was repeated sequentially with a single subgroup being excluded and predictions for that subgroup developed from the other 9, resulting in 10 Cox models. The c-statistic was then calculated for each of the 10 fitting and the prediction data sets.

Back to Top | Article Outline


Baseline Participant Description

At baseline, the 1170 women who had a value of 3 or more on the modified chlamydia risk score6 were predominantly 19 to 24 years of age, black, and of low socioeconomic status (Table 1). Over 15% were nonmonogamous, over 40% had no live births, and about half douched. About 14% reported a prior episode of PID, and almost half reported a history of gonorrhea or chlamydia. Given the study exclusion criteria, no women were pregnant at baseline, none were currently married, and none were virgins.





Enrolled versus eligible, nonenrolled women were more likely to have risk factors for PID, including age <24 years (74.8% versus 67.1%; P = 0.002) and 2 or more sexual partners in the past 2 months (53.8% versus 36.7%; P <0.001). They were less likely to be black (76.1% versus 79.9%; P = 0.001) and did not differ significantly in number of previous pregnancies or douching status. Gonococcal or chlamydial cervicitis was relatively common at baseline, occurring among 14% of women. One hundred one women (8.6%) developed PID.

Individual risk factors related to incident PID (P <0.20) were educational attainment; annual family income; cigarette smoking; history of PID; history of gonorrhea/chlamydia; number of sexual partners in the past 2 months; sex during menses; age at first intercourse; contraceptive use; condom use; sexual partner ever diagnosed with gonorrhea/chlamydia; and baseline infection with N gonorrhoeae, C trachomatis, or both. None of these characteristics had high sensitivity as well as high specificity (Table 2).



In a multivariable Cox regression model including all patient data, 7 dichotomized variables were found to be predictive (P <0.10) of PID outcome: age of first sex 15 or less, current cervicitis, history of PID, annual income <$20,000, current smoking, medroxyprogesterone acetate use, and sex with menses. The area under the ROC, i.e., the proportion developing PID correctly predicted using this prediction model, was 0.74 (95% CI, 0.69–0.78). A model fitted with continuous or interval variables contained the same 7 variables plus history of cervicitis, with the same proportion correctly predicted of 0.74 (95% CI, 0.70–0.78); dichotomizing the age of first sex only and leaving the other variables nondichotomized led to a similar model, with a proportion correctly predicted of 0.74 (95% CI, 0.70–0.78). When only women with biopsy-proven endometritis were analyzed using the final prediction model, the proportion correctly predicted was 0.73 (95% CI, 0.65–0.82). The 10 cross-validation models averaged a proportion developing PID correctly predicted of 0.75 (95% CI, 0.74–0.76) in the fitting subgroups. The average proportion correctly predicted in the 10 excluded subgroups was 0.69 (95% CI, 0.63–0.75); this simulates the accuracy that the model would achieve in an independent but demographically similar group of subjects. Three variables were predictive in all 10 validation models: history of PID, present cervicitis, and current smoking. Medroxyprogesterone acetate use was predictive in 9 of 10 models, and income <$20,000 was predictive in 8. Eight validation models contained either age at first sex or present age; the absence of an age parameter resulted in the largest reduction in the proportion developing PID correctly predicted.

The ROC curves for the original model containing all continuous variables and for the 10 prediction subgroups in cross-validation are shown in Figure 1. ROC curves were determined by making each calculated hazard ratio a cut point and then calculating a sensitivity and specificity for each hazard ratio value. We selected the point at the shoulder of the original curve that maximized sensitivity over specificity as high sensitivity would identify the greatest proportion of at-risk women in whom to focus prevention efforts. Test characteristics were sensitivity 95.3% and specificity 33.4%. Within our high-risk study population, this equated to a positive predictive value of 11.5% and a negative predictive value of 98.7%.

Fig. 1

Fig. 1

Back to Top | Article Outline


Young, single, sexually active women who met a threshold value of 3 or more on a modified Stergachis risk score designed to detect concurrent chlamydial infection developed PID at a rate of 8.6% over a median 3 years of follow-up. This is higher than the 8% lifetime risk of PID reported in a nationally representative sample of American women in 1995.26 It is marginally higher than the PID rates (2.1% in 1 year) reported from the randomized screening trial by Scholes et al.,7 which used the Stergachis model to identify high-risk women participants. We could not develop a model that substantially added to the prediction of the development of PID in this already high-risk cohort. Under a variety of modeling conditions, as few as 69% of women who would ultimately be diagnosed with PID could be correctly predicted. Moreover, picking a point that maximized sensitivity in an attempt to identify the greatest proportion of women who could benefit from more intensive educational and screening interventions would produce an unacceptably low specificity and number of false positives. That is, our analysis validated that a modified chlamydia risk score of 3 or more, which identified women who are sexually active, young, single, and often black, nulligravid, and with prior sexually transmitted infections, was the most efficient predictor for incident PID.

Our observations regarding individual risk factors generally concur with those from numerous retrospective and cross-sectional studies.8–13 In a previous review of risk factors and risk markers for PID, Washington et al.8 summarized that the greatest support was evident for factors including age, socioeconomic status, contraception (barrier, OC, and IUD use), smoking, prompt and compliant treatment of PID episodes, and possibly douching. Our multivariable analyses consistently confirmed the risk associated with age, current smoking, and prior episodes of PID. Not surprisingly, baseline gonococcal or chlamydial cervicitis also predicted PID in our models. Although OC use did not enter the multivariable model, medroxyprogesterone acetate use was an internally consistent protective factor. This raises the possibility that the protection for PID previously observed with OC use and attributed to a diminution of inflammation may be a function of progesterone.20 Previous retrospective and cross-sectional studies have linked smoking to PID,27–29 although confirmatory prospective data have not previously been available. Smoking is thought to exert a biologic effect on PID via compromised immunity or altered estrogen status. Alternatively, smoking may mark poor health-seeking behavior.5 The relationship between medroxyprogesterone and PID has rarely been examined. In a previous ana-lysis from a study of women with prevalent PID, we did not find an association between medroxyprogesterone use and upper genital tract gonorrhea or chlamydia infection.29

The 2 factors that we could not confirm as elevating PID risk were douching and lack of condom use. We previously conducted a detailed analysis of douching as a risk factor for the acquisition of PID in this cohort and found that douching (either at baseline or within 6 months of the outcome) did not increase the occurrence of PID.30 Moreover, in a prospective clinical trial by Rothman et al.,31 douching did not elevate the risk of incident PID. Previous studies linking douching to PID have been retrospective and thus could not establish that douching preceded PID.32 The lack of association between consistent condom use and PID is difficult to explain as it contrasts with the results of several previous studies, including our own.33–35 Notably, incorrect recall of condom use is a well-known problem that would be expected to dampen the observed protective effect.36 Bacterial vaginosis also did not enter the multivariable model as a significant predictor. This is consistent with and discussed in detail in our previously published analysis from this same, prospective GIFT study.17

Strengths of our study include the large number of women studied, use of consistent enrollment and data-collection protocols, collection of biomarkers of effect, relatively long-term and complete longitudinal data collection, and application of survival-based multivariable modeling techniques.

Weaknesses of our study are as follows. The self-reported nature of most of the studied risk factors would have likely underestimated the power of the prediction rule. The observational nature of the study, with the attendant possibility of confounding, would be a weakness in a risk-factor study but is not a weakness in a risk-prediction study, because it is the variance within the data captured by risk factors and not the causal nature of relationships that is important. Screening and timely treatment of cervicitis may have diminished the observed predictiveness of risk factors for PID. That is, gonococcal or chlamydial cervicitis may have been less predictive of subsequent PID than they would have been in an untreated population. Moreover, the infrequency of follow-up interviews, which occurred every 6 months, may have led to an underestimate of the incidence of PID as episodes that occurred between follow-ups may have been underreported. Nonetheless, women in the study experienced high rates of PID.

Another potential weakness is the diagnosis of PID, which was based on clinical, as well as histologic, criteria. We have previously demonstrated that endometritis is no better a predictor of sequelae after PID than are the clinical criteria used in this study, leading us to conclude that endometrial biopsy is not necessary in making the diagnosis of PID.37 Moreover, in sensitivity analyses, we limited our outcome to PID based on a diagnosis of endometritis, and this had little effect on our results.

The study population consisted of women at high risk for developing PID, as evidenced by an 8.6% PID incidence rate. Demographically, they were young, single, and sexually active. They tended to be (but were not uniformly) black, of low socioeconomic status, and with a prior episode of PID or a sexually transmitted infection. These characteristics would describe many populations of women attending STD clinics,38 and so our results would likely apply in such situations. Within our relatively uniformly high-risk population, we would anticipate that a prediction model would have higher predictive values but not a higher likelihood ratio.25 As likelihood ratio was the test upon which our models were based, the results should be valid in higher-risk populations.

Our data do not support the addition of a prediction model to identify a subgroup of women for more intensive intervention within an already at-risk group. Instead, they validate that a modified chlamydia scoring system that results in a group of women (like many seen in STD clinics) who are sexually active, young, single, and often black, nulligravid, and with prior sexually transmitted infections is sufficiently predictive of subsequent PID occurrence that no further risk modeling is necessary to warrant prevention efforts. All such women should undergo intensive prevention education and screening. Moreover, better risk factors and risk markers for diagnosing and predicting PID should be developed and implemented.

Back to Top | Article Outline


1. Lloyd-Jones DM, Wilson PW, Larson MG, et al. Framingham risk score and prediction of lifetime risk for coronary heart disease. Am J Cardiol 2004; 94:20–24.
2. Gail MH, Brinton LA, Byar DP, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. J Natl Cancer Inst 1989; 81:1879–1886.
3. Bach PB, Kattan MW, Thornquist MD, et al. Variations in lung cancer risk among smokers. J Natl Cancer Inst 2003; 95:470–478.
4. Peipert JF, Ness RB, Blume J, et al. Clinical predictors of endometritis in women with symptoms and signs of pelvic inflammatory disease. Am J Obstet Gynecol 2001; 184:856–864.
5. Simms I, Warburton F, Westrom L. Diagnosis of pelvic inflammatory disease: time for a rethink. Sex Transm Infect 2003; 79:491–494.
6. Stergachis A, Scholes D, Heidrich FE, et al. Selective screening for Chlamydia trachomatis infection in a primary care population of women. Am J Epidemiol 1993; 138:143–153.
7. 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; 334:1362–1366.
8. Washington AE, Aral SO, Wolner-Hanssen P, et al. Assessing risk for pelvic inflammatory disease and its sequelae. JAMA 1991; 266:2581–2586.
9. Jossens MO, Eskenazi B, Schachter J, et al. Risk factors for pelvic inflammatory disease. Sex Transm Dis 1996; 23:239–247.
10. Scholes D, Daling JR, Stergachis A, et al. Vaginal douching as a risk factor for acute pelvic inflammatory disease. Obstet Gynecol 1993; 81:601–605.
11. Marks C, Tideman RL, Estcourt CS, et al. Assessment of risk for pelvic inflammatory disease in an urban sexual health population. Sex Transm Infect 2000; 76:470–473.
12. Suss AL, Homel P, Hammerschlag M, et al. Risk factors for pelvic inflammatory disease in inner-city adolescents. Sex Transm Dis 2000; 27:289–291.
13. Miller HG, Cain VS, Rogers SM, et al. Correlates of sexually transmitted bacterial infections among U.S. women in 1995. Fam Plann Perspect 1999; 31:4–9, 23.
14. Hillis SD, Joesoef R, Marchbanks PA, Wasserheit JN, Cates W, Westrom L. Delayed care of pelvic inflammatory disease as a risk factor for impaired fertility. Am J Obstet Gynecol 1993; 168:1503–1509.
15. Kimani J, Maclean IW, Bwayo JJ, et al. Risk factors for Chlamydia trachomatis pelvic inflammatory disease among sex worker in Nairobi, Kenya. J Infect Dis 1996; 173:1437–1444.
16. Baeten JM, Nyange PM, Richardson BA, et al. Hormonal contraception and risk of sexually transmitted disease acquisition: results from a prospective study. Am J Obstet Gynecol 2001; 185:380–385.
17. Ness RB, Hillier SL, Kip KE, et al. Does bacterial vaginosis (BV) increase the risk of incident pelvic inflammatory disease? Obstet Gynecol 2004; 104:761–769.
18. Ness RB, Hillier SL, Richter HE, et al. Douching in relation to bacterial vaginosis, lactobacilli, and facultative bacteria in the vagina. Obstet Gynecol 2002; 765–772.
19. Schwebke JR, Hillier SL, Sobel JD, et al. Validity of the vaginal gram stain for the diagnosis of bacterial vaginosis. Obstet Gynecol 1996; 88:573–576.
20. Ness RB, Keder LM, Soper DE, et al. Oral contraception and the recognition of endometritis. Am J Obstet Gynecol 1997; 176:580–585.
21. Kiviat NB, Wolner-Hanssen P, Eschenbach DA, et al. Endometrial histopathology in patients with culture-proved upper genital tract infection and laparoscopically diagnosed acute salpingitis. Am J Surg Pathol 1990; 14:167–175.
22. McCormack WM, Nowroozi K, Alpert S, et al. Acute pelvic inflammatory disease: characteristics of patients with gonococcal and nongonococcal infection and evaluation of their response to treatment with aqueous procaine penicillin G and spectinomycin hydrochloride. Sex Transm Dis 1977; 4:125–131.
23. Workowski KA, Levine WC. Sexually transmitted diseases treatment guidelines 2002. Centers Dis Control 2002; 51:1–78.
24. Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994; 81:515–526.
25. Grimes DA, Schulz KF. Refining clinical diagnosis with likelihood ratios. Lancet 2005; 365:1500–1505.
26. Abma JC, Chandra A, Mosher WD, Peterson LS, Piccinino LJ. Fertility, family planning, and women's health: new data from the 1995 National Survey of Family Growth. Vital & Health Statistics: Series 23. Data Natl Survey Fam Growth 1997; 19:1–114.
27. Marchbanks PA, Lee NC, Peterson HB. Cigarette smoking as a risk factor for pelvic inflammatory disease. Am J Obstet Gynecol 1990; 162:639–644.
28. Scholes D, Daling JR, Stergachis AS. Current cigarette smoking and risk of acute pelvic inflammatory disease. Am J Public Health 1992; 82:1352–1355.
29. Ness RB, Soper DE, Holley RL, et al. Hormonal and barrier contraception and risk of upper genital tract disease in the PID Evaluation and Clinical Health (PEACH) Study. Am J Obstet Gynecol 2001; 185:121–127.
30. Ness RB, Hillier SL, Kip KE, et al. Douching, pelvic inflammatory disease, and incident gonococcal and chlamydial genital infection in a cohort of high-risk women. Am J Epidemiol 2005; 161:186–195.
31. Rothman KJ, Funch DP, Alfredson T, Brady J, Dreyer NA. Randomized field trial of vaginal douching, pelvic inflammatory disease and pregnancy. Epidemiology 2003; 14:340–348.
32. Martino JL, Vermunol SH. Vaginal douching: evidence for risks or benefits to women's health. Epidemiol Rev 2002; 24:109–124.
33. Eschenbach DA, Harnisch JP, Holmes KK. Pathogenesis of acute pelvic inflammatory disease: role of contraception and other risk factors. Obstet Gynecol 1977; 128:838–850.
34. Kelaghan J, Rubin GL, Ory HW, et al. Barrier-method contraceptives and pelvic inflammatory disease. JAMA 1982; 248:184–187.
35. Ness RB, Randall H, Richter HE, et al. Condom use and the risk of recurrent pelvic inflammatory disease, chronic pelvic pain, or infertility following an episode of pelvic inflammatory disease. Am J Public Health 94:1327–1329.
36. Aral SO, Peterman TA. A stratified approach to untangling the behavioral/biomedical outcomes conundrum. Sex Transm Dis 2002; 29:530–532.
37. Haggerty CL, Ness RB, Amortegui A, Hendrix SL, et al. Endometritis does not predict long-term outcomes after pelvic inflammatory disease. Am J Obstet Gynecol 2003; 188:141–148.
38. Manhart LE, Aral SO, Holmes KK, et al. Influence of study population on the identification of risk factors for sexually transmitted diseases using a case-control design: the example of gonorrhea. Am J Epidemiol 2004; 160:393–402.
© Copyright 2006 American Sexually Transmitted Diseases Association