Comparing Sensitivity, Specificity, Likelihood and Odds Ratios to Predict HIV

The high cost of diagnostic tests in developing countries prompted the adoption of epidemiological tools to predict medical conditions using their clinical signs and symptoms. Tests like Sensitivity (Sn) and Specificity (Sp) have shown to be unreliable as predictors for conditions while the Odds ratio (OR) and Likelihood ratio (LR) proved more robust. The OR and LR have not been used to predict HIV using the World Health Organizations Group I Oral Lesions (GOLs).

Objectives: To compare the Sn, Sp, OR and LR of GOLs for predicting HIV.

Methods: Two Primary Health facilities were randomly selected and a total of 657 patients were screened. The study period comprised of 1 week at each facility.The patients were unaware of their HIV status and underwent a comprehensive oral examination by two calibrated dentists.A saliva HIV test was used to confirm the HIV status. The Sn, Sp, LR and OR for predicting HIV were calculated and compared.

Results: Of the 657 patients that were examined, the HIV prevalence was 34%; the average age was 33.95 (SD=17.82) years and the majority were female (73%). The common lesions were oral candidiasis (46%) and oral hairy leukoplakia (OHL) (19%). Four lesions (angular cheilitis, OHL, pseudomembranous and erythematous candidiasis) were significantly associated with a positive HIV status. The epidemiological tests (Sn, Sp, OR & LR) were confined to these lesions. The Sn was low for single and multiple lesions respectively (22% & 3.4%) while the Sp remained high (94%& 99.8%). Unlike the Sn, the OR and LR for single and multiple lesions (14&12; 30&34) were much more useful as predictors for HIV.

Conclusion: Predicting HIV with GOLs using the sensitivity epidemiological test proved unreliable and inaccurate. The LR and OR proved much more robust in predicting HIV using GOLs.