‘Beyond the Mean’ in Biomarkers Modelling for Economic Evaluations: A Case Study in Gestational Diabetes Mellitus

Economic evaluations often utilise individual-patient data (IPD) to calculate probabilities of events based on observed proportions. However, this approach is limited when interest is in the likelihood of extreme biomarker values that vary by observable characteristics such as blood glucose in gestational diabetes mellitus (GDM). Here, instead of directly calculating probabilities using the IPD, we utilised flexible parametric models that estimate the full conditional distribution, capturing the non-normal characteristics of biomarkers and enabling the derivation of tail probabilities for specific populations. In the case study, we used data from the Born in Bradford study (N= 10,353) to model two non-normally distributed GDM biomarkers (2-hours post-load and fasting glucose). First, we applied fully parametric maximum likelihood to estimate alternative flexible models and information criteria for model selection. We then integrated the chosen distributions in a probabilistic decision model that estimates the cost-effective diagnostic thresholds and the expected costs and quality-adjusted life years (QALYs) of the alternative strategies (‘Testing and Treating’, ‘Treat all’, ‘Do Nothing’). The model adopts the ‘payer’ perspective and expresses results in net monetary benefits (NMB). The log-logistic and Singh-Maddala distributions offered the optimal fit for the 2-hours post-load and fasting glucose biomarkers, respectively. At £13,000 per QALY, maximum NMB with ‘Test and Treat’ (−£330) was achieved for a diagnostic threshold of fasting glucose >6.6 mmol/L, 2-hours post-load glucose >9 mmol/L, identifying 2.9% of women as GDM positive. The case study demonstrated that fully parametric approaches can be implemented in healthcare modelling when interest lies in extreme biomarker values.