Improved screening for pulmonary Kaposi sarcoma in a low-resource setting.

OBJECTIVE

The lung is commonly involved in advanced Kaposi sarcoma (KS) but diagnosis of pulmonary KS in low-resource settings is difficult. Clinical assessments were evaluated to improve screening for pulmonary KS in a low-resource setting.

METHODS

A longitudinal cohort of Africans with HIV-associated KS in the Parirenyatwa Hospital KS clinic, Harare, Zimbabwe, was followed from 2016 to 2018. Assessments included physical exam, chest X-ray, pulmonary function tests, and bronchoscopy with bacterial, fungal and TB cultures. Prevalence of pulmonary KS, tuberculosis (TB), and bacterial or fungal pneumonia were estimated with confidence intervals (CIs). Collective diagnostic performance of combinations of 95 clinical assessments was evaluated by training iterative machine learning models using all available data. Assessments with high importance by cross-validated Shapley value analysis were included in the next iteration. Model performance was evaluated by area under receiver-operator characteristic curves (AUROC).

RESULTS

Among 181 participants with cutaneous KS, 113 had KS lesions in the lower respiratory tract (median 3 anatomic sites; range 1-11 sites). Pulmonary KS prevalence by bronchoscopy was 62.4% (95% CI 55.3, 69.6). Prevalence of TB and fungal/bacterial pneumonia were 9.4% (95% CI 5.3-13.5) and 9.4% (95% CI 5.1-13.7), respectively. A converged set of seven clinical assessments retained in the final model had improved specificity and sensitivity for predicting pulmonary KS compared to individual assessments (AUROC 0.70; 95% CI 0.63-0.78).

CONCLUSION

Pulmonary KS burden remains high in Zimbabwe, despite increased access to antiretroviral therapy. A constellation of clinical assessments was identified to improve screening for pulmonary KS in low-resource settings.