A Parametric Time-to-Event Modelling of Recurrent Ischemic Stroke After Index Stroke Among Patients With and Without Diabetes Mellitus: Implementation of Temporal Validation of the Model

Abstract

Objectives Compared with the first stroke, neurological impairment caused by stroke recurrence is more serious, more difficult to treat, and has a higher mortality rate, especially among ischemic stroke (IS) patients with diabetes mellitus (DM). Although there are established correlations between factors and IS recurrence, there were some issues regarding the naive hazard of IS recurrence with no risk factor influence, and how does the baseline hazard differ among patients with DM and non-DM? To answer all these questions, two time-to-event (TTE) models of recurrent IS after the index IS were developed among IS patients with DM and non-DM. Method A total of 7697 patients with an index IS attack were extracted from the Malaysian Registry of Neurology and stratified according to DM status. Several parametric survival models were evaluated using nonlinear mixed-effect modeling software (NONMEM 7.5). The final model was determined according to the lowest objective function value, graphical evaluation, numerical diagnostics, and clinical plausibility. Additionally, the final model was validated internally and temporally using Kaplan-Meier visual predictive checks (KM-VPCs). Results One hundred ninety-five (5.82%) of 3493 DM patients and 138 (3.28%) of 4204 non-DM patients developed a recurrent IS with a maximum follow-up of 7.37 years. Gompertz's model best fitted the data. With no influence on risk factors, the index IS attack was predicted to contribute to the hazard of recurrent IS by 0.356 and 0.253 within the first six months after the index IS among patients with and without DM, respectively. Even after six months of index IS, the recurrent IS baseline hazard was not equal to zero among both groups (0.0023, 0.0018). Moreover, after incorporating the time and risk factors, the recurrent hazards increased exponentially during the first three years after the index IS followed by an exponential reduction afterward. The recurrent IS predictors among DM patients were ischemic heart disease (IHD) and hyperlipidemia (HPLD). IHD and HPLD increased the hazard of recurrent IS by 2.40 and 1.88 times, respectively, compared to those without IHD and HPLD before index IS (HR, 2.40 (1.79-3.20)), and (HR, 1.88 (1.44-2.45)) respectively. Conclusion The baseline hazard was the highest during the first six months after the index IS. Moreover, receiving medications for secondary prevention failed to demonstrate a significant association with reducing IS recurrence among IS patients with DM, suggesting a need for more intensive patient screening and new strategies for secondary prevention among IS patients with DM.

Keywords: diabetes; ischemic stroke; nonmem; recurrent ischemic stroke; time to event.

Figure 1. Baseline hazard among patients with i) DM and ii) non-DM during and after the first six months of ischemic stroke

Figure 2. Significant covariates influence the hazard of recurrent IS among patients with (left panel) and without (right panel) DM.

IHD: ischemic heart disease; HPLD: hyperlipidemia; HTN: hypertension

Figure 3. Kaplan-Meier VPCs for i) patients with diabetes (probability of not having recurrent ischemic stroke among patients with diabetes), ii) patients with non-diabetes (probability of not having recurrent ischemic stroke among patients with non-diabetes), throughout different time intervals

Figure 4. . Kaplan-Meier plots for temporal validation of the final developed model among diabetic patients throughout different time intervals