Interpreting the model: a systematic review of health outcomes and risk metrics in temperature–health research

While the epidemiological literature on extreme temperatures and health has expanded rapidly, limited attention has been paid to how models operationalise and report outputs - namely, the selection of health outcomes, the application of stratification practices, and the use of risk metrics. These modelling choices are critical for assessing vulnerability and informing equitable climate adaptation strategies, yet remain inconsistently reported and poorly standardised across studies. We conducted a systematic review of 181 peer-reviewed studies published between 2014 and 2026 that model the health impacts of extreme temperatures. Following PRISMA guidelines and building on a companion article focused on model inputs, this review examines how model outputs are specified and communicated. We classify outcome types (e.g. all-cause vs. cause-specific mortality), stratification variables (e.g. age, sex, socio-economic status), and risk metrics (e.g. relative risks, odds ratios, attributable fractions), assessing their prevalence, consistency, and methodological implications. We find wide heterogeneity in output structures. Most studies rely on all-cause mortality, with limited use of diagnostic granularity. Age and sex are the most common stratification axes, while socioeconomic and spatial stratifications remain underused. Risk communication is similarly diverse, with relative risks dominating but often lacking contextual comparability due to differing reference definitions. Few studies report multiple output types or assess robustness across stratification schemes. Current modelling practices inadequately capture the complexity of thermal vulnerability. The absence of standardised output frameworks limits comparability, hinders cumulative knowledge, and risks masking structural health inequalities. We call for clearer reporting standards and propose a core output specification to support transparency and policy relevance in future research.