The World Health Organization (WHO) estimated that globally there were 333 million cases of sexually transmitted infections (STIs) in 1995.1 Most of these infections occurred among women of reproductive age.2 Untreated chlamydial and gonococcal infections are associated with pelvic inflammatory disease, ectopic pregnancy, infertility and neonatal infection. Evidence from a randomized, controlled trial in Tanzania suggests that treatment of STIs can decrease the incidence of HIV infection.3 In response to the HIV epidemic, calls were made at the 1994 International Conference on Population and Development in Cairo and at the Fourth World Conference on Women in Beijing in 1995 for the integration of STI prevention and treatment services into existing family planning and maternal and child health programs.4

Just as all family planning service delivery points do not offer all contraceptive services, they may not be able to offer the entire range of STI services, from prevention counseling and promotion of simultaneous barrier and nonbarrier method use for dual protection to STI management and partner notification. Policymakers and family planning program managers are responsible for making decisions about program expansion according to the STI prevalence among their family planning clients and the availability of resources for prevention, diagnosis and treatment. Mechanisms that identify clients at greatest risk of having an STI would help programs optimize the use of their STI prevention and care resources.5

There is a considerable body of research on STI risk factors. Studies suggest that the usefulness of these risk factors for the purposes of STI screening varies and is strongly dependent on the prevalence of infection and on other characteristics of a given population.6 Most studies among family planning clients suggest that risk factors are not sufficiently predictive to be useful as STI screening tools.7 This may be due in part to a relatively low prevalence of infection, as well as a woman's risk of infection being less dependent on her behavior than on that of her partner, which is difficult to evaluate.8 In settings with higher STI prevalence, overtreatment resulting from the use of risk factors or other nonspecific presumptive treatment tools might be considered an acceptable sacrifice for the sake of reducing STI transmission.9

Syndromic management of STIs uses algorithms based on common signs and symptoms to guide presumptive treatment. Use of a vaginal discharge algorithm for women is limited by its dependence on symptoms. Many women infected with STIs are asymptomatic, and among symptomatic women, signs and symptoms are often nonspecific and may not be related to sexual behavior.10 Researchers have tried to improve the performance of vaginal discharge algorithms among women by adding social and demographic risk characteristics to create algorithms with higher sensitivity and specificity.11

The analysis described in this article was undertaken to identify risk factors for common STIs and to investigate the accuracy of STI algorithms among a sample of family planning clients in Jamaica. A previously published analysis indicated that 27% of these women had an STI.12 Building on this previous research, we used multivariable statistical analyses to reevaluate STI risk factors and to create and assess modified decision models. The goal is to develop practical, efficient and effective strategies to identify women with STIs during routine family planning visits, in order to offer prevention information, additional STI education, condoms and some form of STI management or referral services, thereby optimizing each clinic contact.


Data Collection

A cross-sectional survey was used to collect information on STI prevalence and risk factors from clients of one public and one private nonprofit family planning clinic in Kingston, Jamaica. Women between 18 and 49 years of age* who were not pregnant and who had been using a family planning method during the six months prior to the study were eligible to participate in the survey. Informed consent was required for enrollment.

The women were interviewed about their social and demographic characteristics, STI history, sexual behaviors and contraceptive use. Nurse clinicians were trained to perform pelvic examinations and collect specimens to assess the acidity (pH) of vaginal fluids and confirm the presence of Neisseria gonorrhoeae (using a modified Thayer-Martin culture [BBL; Becton-Dickinson, Cockeysville, MD, USA]), Chlamydia trachomatis (using an enzyme-linked immunoassay [Baxter Bartels, Dundee, Scotland]) and Trichomonas vaginalis (using an InPouch culture [Biomed Diagnostics, San Jose, CA, USA]). A laboratory assistant tested urine samples using a leukocyte esterase dipstick (Ames, Chicago, IL, USA). The dipstick changes color, depending on the concentration of white blood cell enzymes in the urine (on a scale of negative, trace, 1+, 2+, 3+), thus indicating the likelihood of an infection. Women identified as infected by means of a vaginal discharge algorithm (a diagnostic and treatment algorithm based on sexual history, symptoms and clinical signs, as outlined in Figure 1) were given free treatment for gonorrhea, chlamydia or trichomoniasis at the initial visit. All women were asked to return in seven days for the results of laboratory testing. Those with positive laboratory results but who had not received medication were treated at the follow-up visit. Women who received treatment were also given free medication for their partners.

Statistical Methods

Questionnaire data were double-entered and verified using Epi Info 6.0 (Centers for Disease Control and Prevention, Atlanta, GA, USA), and analyzed using SPSS 6.1 for Windows (SPSS Inc., Chicago, IL, USA) and SAS 6.11 for Windows (SAS institute, Cary, NC, USA). To determine individual factors related to infection, we calculated odds ratios and 95% confidence intervals for each potential risk factor. Using multiple logistic regression, we analyzed all factors found to be significant through bivariate analysis (p<.05). In addition, we included other commonly recognized risk factors, such as having a partner with other partners or a partner with urethral discharge, which were not statistically significant in bivariate analysis. We excluded variables that were significant in bivariate analysis from the multiple logistic regression if they were highly correlated with other independent variables that were more inclusive (those with Spearman correlation coefficients of 0.3 or more).

When gonorrheal, chlamydial or trichomonal infections are symptomatic, they are often characterized by an abnormal vaginal discharge. In this analysis, we therefore grouped the three infections as one outcome variable. In a separate analysis, cervical infection caused by N. gonorrhoeae or C. trachomatis was considered a dependent variable because of the serious sequelae associated with untreated cervical infection.

We evaluated our decision models for their ability to correctly identify women with gonorrhea, chlamydia or trichomoniasis, based on measures of sensitivity, specificity and positive predictive value. (Positive predictive value is defined as the proportion of women with infections confirmed by laboratory analysis out of the women identified or selected by the decision model.) Models were compared with a WHO-based algorithm, modified for use in Jamaican STI clinics (Figure 1).13 We did not separately analyze algorithms among the subset of women who reported a discharge, since vaginal discharge was not significantly associated with cervical infection or trichomoniasis. All women, regardless of symptoms, are included in the analyses. Decision models for syphilis were not evaluated, since simple laboratory screening for syphilis among low-risk women has been shown to be cost-effective.14

The two weighted scoring models are based on risk scores assigned to each significant risk predictor. Similar to methods used by Vuylsteke,15 we multiplied the coefficients of significant covariates (identified through logistic regression) by 10. A woman's risk score is the sum of the products derived from each covariate. Women are categorized as infected with an STI if their total score is greater than or equal to a given cut-off score (nine or more in Model 1, and eight or more in Model 2). Reported spotting or bleeding after sex may be clinically related to a friable cervix (one that easily bleeds upon contact, e.g. with a cotton swab); thus, to simplify the model, we removed cervical friability from the second model and substituted postcoital spotting, thereby eliminating the need for a pelvic examination.

The rapid risk-assessment model is based on six risk factors that were either confirmed through multiple logistic regression analysis (positive urine dipstick result, more than one partner in the past year, younger than 25 and spotting or bleeding after sex) or other commonly recognized factors (partner who has other partners and partner who has urethral discharge).16Again, we replaced cervical friability with postcoital spotting. A woman with two or more risk characteristics was classified as infected. For comparison, we also present a model including interview information alone (designated as "risk questions"); this decision model is identical to the rapid risk assessment without the urine test.


A total of 782 female family planning clients were recruited from June to November 1995. Fifteen women who completed the risk-factor questionnaire were excluded for failure to complete the medical examination, leaving 767 (98%) who were included in the analysis.

Multiple logistic regression (p<.05) identified four significant risk factors for gonococcal, chlamydial or trichomonal infection: having a urine dipstick result greater than 1+, having had more than one partner in the past year, having a friable cervix on examination or being younger than 25 (Table 1). Factors found to be significant only through bivariate analysis included spotting after sex, vaginal odor, more than one partner in the past three months and a casual partner or a partner who has other partners.

Multivariable analysis identified urine dipstick results greater than 1+, more than one partner in the past year, a friable cervix on examination, being younger than 25 years and spotting after sex as risk factors for cervical infection alone (gonococcal or chlamydial). Bivariate analysis suggests that having had more than one partner in the past three months was also related to cervical infection (Table 2, page 204).

In identifying gonorrhea, chlamydia or trichomoniasis in this sample of family planning clients, the WHO-based risk-inclusive algorithm modified for Jamaican STI clients (Figure 1) was 58% sensitive, was 46% specific and had a positive predictive value of 25% (Table 3, page 205).17 In comparison, the two weighted-risk algorithms (Models 1 and 2) and the rapid assessment model were more accurate, with positive predictive values of 39%, 38% and 35%, respectively. The risk questions alone outperformed the WHO-based algorithm in terms of positive predictive value (34%), although this approach was slightly less sensitive (54%) than the others (58-71%).

The relative accuracy of these approaches for identifying cervical infection but not trichomoniasis was similar to that found for all STIs. The modified WHO risk-inclusive algorithm was least effective in assessing the likelihood of current gonococcal or chlamydial infections based on the positive predictive value (14%). The weighted-risk algorithms had the highest positive predictive values (23%), while the rapid risk assessment and the risk questions were slightly less predictive than the weighted models (20% each) but more so than the WHO algorithm.


Health policymakers are faced with the growing STI problem among populations traditionally considered at lower risk. Almost all of the participants in this study were family planning clients who presented for routine contraceptive management; however, more than one-quarter were diagnosed with at least one of four STIs. Most women were asymptomatic or did not recognize their symptoms as abnormal and potentially treatable until questioned. Even among women who reported vaginal discharge, this symptom was not a significant predictor of cervical infections or trichomoniasis. These findings and others suggest that it is often difficult to correctly identify women with STIs based on symptoms.18

Our original goal was to assess STI prevalence and to identify risk factors among family planning populations in Jamaica. These individual factors are the basis for decision models designed to identify infected women. We tested the Jamaica STI clinic algorithm (i.e., the modified WHO algorithm) for its usefulness among this lower-risk group. Finding it an ineffective STI management tool among asymptomatic and symptomatic family planning clients, we attempted to develop acceptable alternative decision models to identify women at highest risk of current infection during routine family planning visits. Ideally, decision models could be used to focus STI prevention activities, to select or defer IUD candidates, to offer presumptive treatment for the patient and the patient's partner, or to refer them for further evaluation and treatment.

Multivariable analysis identified several individual predictors of infection that have been significant in other research. A urine dipstick reading greater than 1+ was significantly associated with cervical infections or trichomoniasis, and was the strongest predictor in the cervicitis and vaginitis model. Other research suggests varying degrees of association between a positive urine dipstick result and infection among women. For detection of gonorrhea, one study among female STI clients found the urine dipstick to be as accurate as Gram stains.19 For detection of gonorrhea or chlamydia, the same study found the urine dipstick to be as sensitive as, but less specific than, Gram stains. It also was predictive of cervical infection among antenatal study participants in Zaire,20 while in Kenya the urine dipstick predicted vaginitis but not cervical infection.21 It is relatively simple and inexpensive to administer and may prove useful as an adjunct screening tool for STI counseling and management.

Consistent with findings of several studies among antenatal and family planning clients in various settings, we found that having more than one partner in the past year was also a significant predictor of cervical infection and trichomoniasis in multivariable analysis.22 Among U.S. family planning clients and among antenatal clients worldwide, age is the best predictor of chlamydial infection, with younger age-groups being at higher risk of infection.23 Although less commonly evaluated, cervical friability was identified as a predictor of cervical infection or trichomoniasis in this and other developing countries.24 Several other STI risk factors identified in other research were not found to be significant here: being unmarried or single;25 currently using injectable hormones, oral contraceptives or IUDs;26having a vaginal discharge;27 experiencing vaginal itching;28 experiencing cervical motion tenderness;29 or having a partner with urethral discharge.30

In the analysis of the data collected in Jamaican family planning clinics, we evaluated algorithms that included demographic and behavioral risk factors for their ability to identify women with gonorrhea, chlamydia or trichomoniasis. Decision models for the management of any one of the three infections (rather than cervical infection alone) might be used for counseling and referral, but not for presumptive treatment. Consistent with findings of other research,31 the WHO modified risk-inclusive algorithm was the least predictive. The proportion of infected women among those selected by the algorithm was comparable to the prevalence of infection in the entire study population; thus, it performed no better than chance. Moreover, this approach is the most complicated to use. Results here suggest that it is an inappropriate screening tool for women who are not already seeking STI services.

It should be emphasized that the WHO STI algorithm was not developed for use with asymptomatic clients (men or women). The WHO recognizes that the algorithms are not necessarily applicable across regions and recommends that syndromic approaches be tested and modified among local populations before being implemented. Among symptomatic clients seeking care, decision models that rely on a physical examination have been shown to be useful to identify pelvic inflammatory disease, epididymitis, genital ulcer syndrome and bacterial vaginosis.32

In this article, the weighted-risk models were more predictive than the modified WHO algorithm; however, Model 1 requires a physical examination, and both weighted models rely on calculation of a risk score. Experience has shown that providers do not always accept the use of risk scores, and that calculation may sometimes be viewed as complicated and bothersome. The accuracy of the rapid risk assessment was similar to the weighted-risk algorithms, with several advantages. This simple tool requires minimal training and inexpensive supplies. No clinical examination is required, and staff need not calculate risk scores. Training, facilities, equipment and supplies required for pelvic examinations (electricity, sterilizer, examining tables, lamps, specula and drapes) are unnecessary. Even without access to urine dipsticks, providers may find the risk questions alone preferable to the WHO algorithm for identifying family planning clients in need of STI counseling, referral or presumptive treatment.

Few family planning programs in developing countries have the human resources or capital to enable staff to perform hygienic and effective pelvic examinations in settings other than scarce private or hospital-based clinics. The high level of asymptomatic infection remains an STI management dilemma for clinicians faced with limited access to diagnostic laboratory tests. A recent review of non-laboratory-based decision models suggests that they are generally not sensitive or specific enough to be used for identification of infection among antenatal and family planning populations.33 The results of our analysis support these assertions.

Among the decision models evaluated for detection of cervical infection, the highest positive predictive value was only 23% (weighted scoring). Other studies suggest that among populations with a lower prevalence of infection, decision models for screening or management of cervical infection may result in positive predictive values even lower than those here.34 This should serve to remind program managers of the importance of considering the prevalence of a particular infection in a given population before they implement decision models. Nevertheless, as mentioned above, in order to curb STI transmission among family planning populations with high STI prevalence, overtreatment resulting from use of risk factors or other nonspecific presumptive treatment tools might be justified,35 particularly given that there are single-dose treatments that are relatively inexpensive, safe and easy to administer.

For effective control, the prevention and management of STIs must become a higher priority for health care providers serving sexually active populations who were previously assumed to be at lower STI risk.36 To optimize each contact, providers should offer women prevention information and condoms. Risk evaluation, presumptive treatment, etiologic diagnosis and treatment, or referral services should be provided to the extent that resources allow in areas where STI prevalence may be high.

Reliable and economical STI classification techniques that do not require physical examinations are urgently needed. Non-laboratory-based decision models can be used to select or refer women for further evaluation and treatment, or to offer presumptive treatment where resources allow, where prevalence suggests or when the patient is unlikely to seek further evaluation.

Until methods such as DNA amplification techniques are globally available, the rapid risk assessment and other STI decision models should be evaluated in semiurban and rural sites in Jamaica or other settings with similar populations. Economic analysis of different management strategies will also enable health policy officials to make more informed choices among alternative approaches.