Background: Obtaining CD4⁺ cell counts are expensive and not practical for routine monitoring in many resource-limited settings. Total lymphocyte count (TLC) has been used as a surrogate CD4 marker but with varying sensitivities and specificities.

Methods: Patients were drawn from an ongoing, observational database cohort study at the UAB HIV clinic. Patients were included if they had CD4⁺ cell count and Complete Blood Count (CBC) (TLC, hemoglobin, and platelet count) drawn on the same day. Only one randomly selected set of laboratory data was used per patient. The remaining data were used to validate the models generated. Models predicting CD4+ cell count £ 200/μL were developed by 1) decision tree analysis, and 2 )multivariable logistic regression analysis. The variables included in the analyses were TLC, hemoglobin, platelet count, gender, and any antiretroviral therapy (ART) in the previous 30 days (yes/no).

Results: We studied 1,189 patients (pts). Median age was 38 yrs, 55% were Caucasian, 77% male, median CD4⁺ cell count was 333/μL, 65% were on ART. Overall sensitivity, specificity, and positive predictive value (PPV) of the decision tree to predict CD4⁺ cell count £ 200/μL were 89%, 77%, and 89%, respectively. Gender and ART were not significant in the decision tree analysis. Multivariable logistic regression analysis achieved similar results. The final multivariable model for predicting CD4⁺ cell count £ 200/μL included TLC, hemoglobin, platelet count, gender, gender*hemoglobin, and gender*platelet count, with sensitivity = 91%, specificity = 73%, and PPV = 88%.

Conclusion: Using inexpensive laboratory values of the CBC, the decision tree analysis provides an effective model that is much simpler to use than a logistic regression equation to predict whether the CD4+ cell count is £ or > 200/μL. This decision tree has immediate relevance and applicability to resource-limited settings.