What this is

A post hoc subgroup analysis of the LAMP trial examining whether baseline insulin resistance (IR) modifies the effect of liraglutide — a GLP-1 receptor agonist started within 24 hours of acute minor stroke or high-risk TIA — on recurrent stroke in patients with type 2 diabetes. The central finding: liraglutide delivered dramatic benefit only in patients with HOMA-IR ≥ 2.5, while patients with lower IR saw no significant reduction in stroke recurrence. This reframes the GLP-1 RA stroke story from a blunt population-level intervention to a precision-medicine stratification opportunity.

1. Shadow Audit

The shadow audit asks: what is the hidden structural feature in this dataset that most people miss?

The answer is the control group IR paradox. In the control arm, patients with IR had higher stroke recurrence (18.1%) than non-IR patients (9.5%), with HR 2.00 (95% CI 0.92–4.34, P = 0.08) and functional disability (mRS > 1) at 30.7% vs 13.4% (OR 2.86, 95% CI 1.45–6.08, P = 0.004). This means IR is an independent amplifier of stroke risk — it is not merely a marker but a causal force. Yet in the liraglutide arm, IR patients dropped to 5.8% stroke recurrence, while non-IR patients rose to 11.5%. The drug didn’t just neutralize IR’s harm — it inverted the hierarchy entirely. The shadow is that without liraglutide, IR patients face roughly 2× the stroke risk; with liraglutide, they become the best responders (HR 0.50, 95% CI 0.20–1.23).

2. Inversion Engine

The inversion engine identifies the core finding that, if reversed in understanding, yields the opposite but equally valid insight.

Conventional reading: liraglutide works better in insulin-resistant patients. The inversion: liraglutide may not work at all in non-IR patients. Among the 162 patients with HOMA-IR < 2.5, stroke recurrence was 11.5% in the liraglutide arm versus 9.5% in controls — an absolute risk increase of +2.0% (HR 1.22, 95% CI 0.47–3.16, P = 0.42). That’s not null — it’s a point estimate trending toward harm. The treatment × IR interaction was significant at P = 0.02 for both stroke and composite vascular events. This inversion matters clinically: giving liraglutide to the roughly 31.8% of T2D stroke patients with preserved insulin sensitivity may be wasted pharmacotherapy, and the restricted cubic splines show a nonlinear interaction (P for nonlinear interaction = 0.02), suggesting the treatment effect curves cross at specific HOMA-IR thresholds.

3. Second-Order Catalyst

What downstream effect does this paper catalyze beyond its immediate conclusions?

The second-order catalyst is IR-guided acute stroke pathway redesign. The study showed liraglutide benefit emerged as early as day 12 (HR 0.511, 95% CI 0.263–0.995, P = 0.048) in the overall population, and in the IR subgroup, statistical significance was first achieved at day 5. If IR-stratified protocols are adopted, hospitals could triage acute T2D stroke patients by a simple bedside HOMA-IR calculation (FPG × FINS / 22.5) within the 24-hour treatment window. This would redirect liraglutide prescriptions — currently given to all eligible T2D stroke patients — toward the 68.2% who actually benefit, while sparing the 31.8% with preserved insulin sensitivity. The NNT of 8 in the IR group versus no measurable benefit in non-IR makes this one of the starkest stratification signals in recent stroke pharmacology.

4. Asymmetric Leverage

Where does a small input produce an outsized output?

The asymmetric leverage lies in the NNT = 8 for both stroke recurrence and composite vascular events in the IR subgroup. The absolute risk reduction was 12.3% (95% CI 5.6%–19.0%) for stroke and 13.4% (95% CI 6.6%–20.2%) for vascular events. For context, the original LAMP trial showed an overall HR of 0.48 for stroke recurrence (7.6% vs 15.3%). But the IR subgroup alone drove the entire signal: 5.8% vs 18.1% (HR 0.30, 95% CI 0.15–0.62). The small input — a single HOMA-IR calculation at admission — produces the outsized output of identifying a population where liraglutide converts from marginally effective to remarkably potent, while also revealing that HOMA-IR did not change significantly at week 1 (geometric mean ratio 1.08, 95% CI 0.95–1.22, P = 0.26), suggesting the mechanism is not simply metabolic correction.

5. Paradigm Destroyer

What existing paradigm does this evidence dismantle?

This paper attacks two entrenched paradigms simultaneously. First, the weight-loss-as-mechanism narrative for GLP-1 RA cardiovascular benefit. The exploratory analysis showed BMI < 24 kg/m² patients had HR 0.30 (95% CI 0.12–0.74, P = 0.008) while BMI ≥ 24 had HR 0.55 (95% CI 0.28–1.11, P = 0.35), with no significant categorical interaction (P = 0.15) but a significant nonlinear interaction in restricted cubic splines (P = 0.006). The leanest patients had the largest benefit — the opposite of what a weight-loss mechanism would predict. Second, it destroys the uniform-benefit assumption. The control group showed a U-shaped BMI–recurrence curve (mirroring the obesity paradox), yet IR — not BMI — was the true effect modifier. hs-CRP interaction was not significant (P = 0.40), meaning even inflammation didn’t stratify response as cleanly as IR. The paradigm that GLP-1 RAs work equally across metabolic phenotypes is broken.

MVP — Minimum Viable Proof

What is the smallest, most compelling evidence package that proves the core claim?

The MVP is a single table: Table 2 from the paper, showing treatment effects stratified by IR status. In the IR subgroup (n = 348, 68.2% of the cohort), liraglutide reduced stroke recurrence from 18.1% to 5.8% (HR 0.30, 95% CI 0.15–0.62). In the non-IR subgroup (n = 162), the rates were 9.5% vs 11.5% (HR 1.22, 95% CI 0.47–3.16). The interaction P = 0.02. This is confirmed by multiple imputation analysis (interaction P = 0.025) and sensitivity analysis using HOMA-IR ≥ 2.0 (consistent primary results). The Kaplan-Meier curves (Figure 3) visually demonstrate separation beginning early and persisting through 90 days, with Schoenfeld residuals confirming proportional hazards (global P = 0.28, non-IR subgroup P = 0.80, IR subgroup P = 0.36).

Best Combination

What is the optimal pairing of insights from this paper?

The best combination is IR-stratified initiation + early time-to-benefit evidence. The paper demonstrates that in IR patients, liraglutide’s statistical significance was first reached at day 5 (Figure 4A), while in non-IR patients, the minimum P value was at day 49 and never achieved significance (Figure 4B). Combined with the overall early benefit emerging at day 12 (HR 0.511, P = 0.048) and persisting at 30 days (HR 0.448, P = 0.0095), 60 days (HR 0.518, P = 0.0199), and 90 days (HR 0.478, P = 0.0082), the optimal protocol is: measure HOMA-IR at admission, initiate liraglutide within 24 hours in IR patients, and expect meaningful risk reduction within the first week. The 0.6 mg → 1.2 mg → 1.8 mg/d titration schedule means even at subtarget doses, the effect begins early — suggesting that the dose-response relationship may have a steep initial slope in metabolically primed patients.

Overvalue Warning

Where are readers most likely to overinterpret or over-rotate on this evidence?

Three critical overvalue warnings. First, this is post hoc, not prespecified — the authors explicitly state this was not in the original protocol, with no correction for multiple comparisons, increasing type I error risk. Second, HOMA-IR only reflects hepatic insulin resistance and may not capture peripheral (skeletal muscle) IR, meaning the 31.8% classified as non-IR may include patients with occult systemic dysfunction — the authors note that two-thirds of non-IR patients had prior glucose-lowering therapy, which may have normalized fasting indices despite persistent metabolic disease. Third, the population is exclusively Chinese (27 hospitals in China), and HOMA-IR cutoffs vary by ethnicity — the 2.5 threshold used here is validated in Asian populations but may not generalize. Sensitivity analyses with HOMA-IR ≥ 2.0 supported robustness (P = 0.02 for primary interaction), but the fundamental question of whether these findings replicate in European or African-descent populations remains open.