Personalised risk-prediction tools for cryptococcal meningitis mortality to guide treatment stratification in...

Cryptococcal meningitis is a major driver of global HIV-related mortality, and validated approaches to stratify mortality risk could help to target effective treatment strategies. We aimed to develop and validate models to predict risk of all-cause mortality in people with HIV-associated cryptococcal meningitis in sub-Saharan African countries.

For this prediction modelling study, we pooled individual-level data from the ACTA (ISRCTN45035509) and AMBITION-cm (ISRCTN72509687) randomised controlled trials. Data in ACTA were collected between Feb 12, 2013, and Jan 10, 2017, and data in AMBITION-cm were collected between Jan 31, 2018, and June 11, 2021. Adults aged 18 years or older with a first episode of HIV-associated cryptococcal meningitis were recruited to both trials. Exclusion criteria included pregnancy or lactation; receipt of high-dose anti-fungal treatment doses before screening; and contraindications to trial medication. Participants were recruited from nine hospitals across Cameroon, Malawi, Tanzania, and Zambia in ACTA and eight hospitals across Botswana, Malawi, South Africa, Uganda, and Zimbabwe in AMBITION-cm. We developed two primary multivariable logistic-regression models for the primary outcome of 2-week mortality: a basic model for use in a resource-limited setting that contained only candidate predictors that are routinely, programmatically obtained at hospital admission and a research model for which all predefined candidate predictors were considered for inclusion. We used internal-external cross-validation to evaluate model performance across countries within the development cohort (ie, data from all countries except Malawi participants in AMBITION-cm), before validation of discrimination, calibration, and net benefit in held-out data from Malawi.

We included 674 eligible participants from ACTA and 814 from AMBITION-cm in the pooled analysis (total sample size 1488). 1263 participants were included in model development, with 225 from the Malawi site in AMBITION-cm held out for validation. 222 (17·6%) of 1263 participants in the development set and 21 (9·3%) of 225 participants in the validation set met the primary model outcome of 2-week mortality. We retained five predictors in the basic model and seven in the research model. Predictors in both models were Glasgow Coma Scale score, Eastern Cooperative Oncology Group performance status, haemoglobin, blood neutrophil count, and treatment. Additional predictors in the research model were cerebrospinal fluid opening pressure and log10 cerebrospinal fluid quantitative cryptococcal culture. Discrimination was relatively consistent between study sites for both models (pooled C statistic 0·75 [95% CI 0·68-0·82] for the basic model and 0·78 [0·75-0·82] for the research model), but calibration was more heterogeneous (pooled calibration slope 0·87 [95% CI 0·57 to 1·17] and 0·83 [0·69 to 0·97], pooled calibration in the large 0·00 [-0·54 to 0·55] and -0·02 [-0·46 to 0·42], for the basic and research models, respectively). In held-out validation, discrimination of both models was slightly higher than estimates from internal-external cross-validation (C statistic 0·78 [95% CI 0·70-0·87] in the basic model and 0·85 [0·79-0·92] in the research model). Calibration assessment suggested overestimation of risk, particularly in the high-risk range: calibration slope 1·04 (95% CI 0·54 to 1·55), calibration in the large -0·55 (-1·02 to -0·07). When comparing single, high-dose liposomal amphotericin B plus 14 days of flucytosine plus fluconazole with 1 week of amphotericin B plus flucytosine in AMBITION-cm, hazard ratios were 0·50 (95% CI 0·26-0·97) in the low-risk stratum and 0·96 (0·67-1·37) in the high-risk stratum for the basic model, and 0·61 (0·31-1·18) in the low-risk stratum and 1·03 (0·72-1·47) in the high-risk stratum for the research model.

Both models accurately predicted 2-week mortality in people with HIV and have the potential to be incorporated into future treatment-stratification approaches in low-income and middle-income countries.

National Institute for Health Research.