Longitudinal trajectories unravel the complex interplay of medication, cardiovascular events, chronic kidney disease, and mortality

Figures

• 

  Fig. 1. The diagram illustrates the data extraction, process modelling and analysis.This figure includes: A flowchart illustrating the selection of new Proton Pump Inhibitor (PPI) and H2 blocker (H2B) users from the SCREAM database. Process model construction and analysis are performed to explore disease trajectories, with statistical tests subsequently used to assess the patterns identified by the process model.

• 

  Fig. 2. A data-driven DAG (directed acyclic graph) synthesises evidence-based insights with data-driven modelling to represent the disease trajectory and guide the data analysis strategy employed in this study. The node represents the event and displays the relative case frequency (in brackets), which is the proportion of individuals (cases) experiencing the event among all included individuals. The edge (arrow) represents time-ordered sequences of events and displays two statistics, relative-antecedent frequency (risk), which is the proportion of exposure event (antecedent) individuals (cases) directly followed by the target individuals (cases); and median time (in brackets) represents the time from the exposure event to the outcome or competing event, in years. Only statistically significant transitions between PPI and H2B, with a transition risk greater than 10%, are shown in the indicator.

• 

  Fig. 3. Comparison of trajectories of new users of PPI or H2B, 2006–2021. Trajectories: Chronological disease trajectories including Medication, CKD outcomes, CVAE, Death/Study end; each trajectory is presented with the relative frequency, absolute frequency and cumulative frequency. Trajectories are ordered by frequency. PPI vs. H2B for the following trajectory: hazard ratios and 95% confidence intervals of undergoing a specific trajectory between PPI and H2B users. Estimates were derived from a Cox proportional hazards model with inverse probability weighting (IPW), adjusting for age, sex, eGFR at index, comorbidities, and concomitant medication use at index.

• 

  Fig. 4. Data-driven DAGs depict the influence of death as a competing risk when evaluating CVAE and CKD outcomes. The node represents the event. The edge (arrow) represents time-ordered sequences of events and displays two statistics, relative-antecedent frequency, which is the proportion of source (antecedent) cases directly followed by the target cases; and median duration (in brackets) from antecedent event to the target event.

• 

  Fig. 5. A data-driven DAG illustrates the mediation framework. The first indirect effect (medication use → CKD) is represented by the association between medication exposure (PPI vs. H2B) and CKD, with the p-value derived from a competing risk model for CKD as the outcome. The second indirect effect (CKD → CVAE) is represented by the association between CKD status (CKD vs. non-CKD) and CVAE, with the p-value obtained from a competing risk model for CVAE as the outcome. The direct effect is represented by the association between medication exposure (PPI vs. H2B) and CVAE with adjusting for CKD status in the competing risk model.