# Prognostic models predicting clinical outcomes in patients diagnosed with visceral leishmaniasis: a systematic review

## TL;DR

This paper reviews existing prognostic models for predicting outcomes in patients with visceral leishmaniasis, highlighting gaps and risks of bias in current models.

## Contribution

The study systematically identifies and evaluates prognostic models for visceral leishmaniasis mortality, revealing evidence gaps and high bias risks.

## Key findings

- Eight studies describing 12 prognostic models for VL mortality were identified, mostly from Brazil and East Africa.
- All models were at high risk of bias due to lack of calibration plots and overfitting from small sample sizes.
- No models predict treatment failure or relapse, and none were developed in high-burden South Asian populations.

## Abstract

Visceral leishmaniasis (VL) is a neglected tropical disease prevalent in populations affected by poverty and poor nutrition. Without effective treatment, death is the norm. Prognostic models can steer clinical decision-making by identifying patients at high risk of adverse outcomes. We aimed to identify, summarise and critically appraise prognostic models predicting future clinical outcomes in patients with VL.

We systematically reviewed all studies that developed, evaluated or updated prognostic models predicting future clinical outcomes in patients diagnosed with VL. Five bibliographic databases (Ovid Embase, Ovid MEDLINE, Web of Science Core Collection, SciELO and LILACS) were searched from database inception to 1 March 2023, with an update to 18 December 2025. Screening, data extraction and risk of bias assessment (Prediction Model Risk of Bias Assessment Tool) were performed independently and in duplicate. Results are presented with tables, figures and a narrative synthesis.

Eight studies, published between 2003 and 2021, were identified describing 12 prognostic model developments. 10 models were evaluated in settings that were either geographically or temporally distinct from those used for model development, resulting in 19 external validations. All models predicted mortality, either using hospital-based cohorts (10 models) or registry data (2 models), and were developed in either Brazilian or East African populations (9 and 3 models, respectively). Model discrimination (c-statistics) ranged from 0.56 to 0.93 when evaluated in the same patients used for model development (apparent performance, 12 models), and 0.62 to 0.92 when evaluated in new settings (19 external validations, 10 models). All model developments and evaluations were judged at high risk of bias: no studies presented calibration plots, 11 models were at high risk of overfitting due to small sample sizes, and four models presented risk scores that did not correspond to the reported regression coefficients.

All identified models predict mortality and were developed in Brazilian or East African patient populations. No prognostic models were identified that predict treatment failure or relapse, and despite South Asia accounting for the highest global VL burden prior to 2010, no models were developed in this population. Within the context of the ongoing elimination programmes in South Asia and East Africa, these represent important evidence gaps where new model development should be prioritised. Information presented in this review can be used by clinicians and policymakers to assess the applicability of existing models to their own patient settings. However, with a high risk of bias identified for all models, caution should be exercised when interpreting model risk and performance estimates. We direct interested readers to expert guidance to support transparent reporting and reduce common sources of bias in the development and evaluation of prediction models.

CRD42023417226.