Why BGPT?
logo

Review papers with raw data transparency

Quickly verify claims by accessing the underlying experimental data and figures.







Press Enter ↵ to solve



    Fuel Your Discoveries




     Quick Explanation


    Paper focus (what they found)

    In a large Taiwan nationwide cohort of adults with T2D + liver cirrhosis, initiating an SGLT2 inhibitor (vs initiating a DPP-4 inhibitor) was associated with lower risks of ESKD, AKI, MACE, and hepatic decompensation, using inverse probability weighting (IPTW) and competing-risk Cox models.



     Long Explanation


    SGLT2 vs DPP-4 in T2D + Cirrhosis: rigorous observational claims, with nontrivial uncertainty

    Study type: Nationwide retrospective cohort using claims/registry data from Taiwan’s NHIRD, comparing initiators of SGLT2 inhibitors vs DPP-4 inhibitors.

    1) Visualizing the effect sizes (adjusted HRs)

    Below are the reported adjusted hazard ratios (SGLT2i vs DPP4i). Hazard ratios are not risk differences; they assume proportionality (or use time-varying coefficients when violations were detected).

    Primary quantitative claims (from the paper)
    • Adjusted HR for ESKD: 0.34 (95% CI 0.25–0.47).
    • Adjusted HR for AKI: 0.66 (95% CI 0.59–0.74).
    • Adjusted HR for MACE: 0.67 (95% CI 0.62–0.71).
    • Adjusted HR for hepatic decompensation: 0.65 (95% CI 0.57–0.74).

    2) Converting incidence rate differences into a “per-1000 PY” view

    The paper reports incidence rates per 1000 person-years and event counts; these can be used to approximate rate ratios (not the same as HRs).

    Rate-level facts from the table
    • ESKD incidence: 7.22 vs 2.30 per 1000 PY (DPP4i vs SGLT2i).
    • AKI incidence: 37.71 vs 22.21 per 1000 PY.
    • MACE incidence: 114.71 vs 65.44 per 1000 PY.

    3) Mechanistic plausibility: what’s supported vs what’s speculative

    What’s plausibly supported by broader evidence (not proved in this paper):
    • SGLT2 inhibitors have established cardiology/nephrology benefits that are commonly attributed to multiple convergent mechanisms beyond glucose lowering (eg, hemodynamic effects, anti-inflammatory/antioxidant pathways, energy substrate shifts, and kidney microcirculation effects).
    • The paper’s liver-focused hypotheses are biologically plausible (shared metabolic/inflammatory pathways; potential effects on NAFLD/MASLD), but the study itself is observational and cannot establish causality.
    What remains uncertain / not directly tested here:
    • No laboratory-based cirrhosis severity (eg MELD-Na, Child-Pugh) was available, so residual confounding by baseline liver function is possible.
    • Confounding by indication: even with IPTW/matching, prescription choice may correlate with unmeasured health status (eg frailty, contraindications, severity of portal hypertension) which could influence both drug selection and outcomes.
    • Outcome misclassification from administrative codes: ESKD/AKI/MACE/hepatic decomp are coded events; misclassification can bias effect estimates. The paper cites validation metrics for some endpoints, but validation doesn’t eliminate bias.

    4) Study design audit: internal validity checks

    Key strengths (why the estimate may be credible):
    • “New-user” style initiation and 365-day washout reduces ambiguity about prior exposure and helps mitigate time-immortal bias by beginning follow-up at index date (t=0).
    • Competing risk handling (cause-specific hazards and cumulative incidence via Aalen-Johansen with competing risk of death) is appropriate given high mortality burden in cirrhosis.
    • Multiple sensitivity analyses (PS trimming, 1:1 matching, outpatient diagnosis inclusion for MACE components, stratification by CKD stage) improve robustness against some modeling choices.
    Most important blind spots (what could still flip the interpretation):
    • Unmeasured liver severity and frailty could drive both treatment selection and event risk; without labs, balancing cannot fully equalize baseline risk.
    • Exposure misclassification**”: pharmacy claims reflect prescriptions, not ingestion/adherence; also switching and off-formulary changes are handled via intention-to-treat on initial assignment (so effect is on initiators, not necessarily sustained dosing).
    • Selection/generalizability: population is Taiwanese; prior viral-hepatitis epidemiology and prescribing patterns may differ elsewhere.

    5) What evidence level should we assign?

    • Known from this paper (data-backed): Observational associations with statistically significant adjusted HRs favoring SGLT2i over DPP4i across multiple endpoints.
    • Inferred but not proven: That the association reflects causal drug effects on disease biology in cirrhosis (rather than residual confounding).


    Feedback:    

    Updated: April 22, 2026

     Top Data Sources ExportMCP


     Science Art


    Paper Review: SGLT2 Inhibitor Use and Cardiorenal Outcomes in Type 2 Diabetes With Liver Cirrhosis Science Art

     Science Movie


    Make a narrated HD Science movie for this answer ($32 per minute)




     Discussion








    Get Ahead With Science Insights

    Custom summaries of the latest cutting edge Science research. Every Friday. No Ads.