Cohort studycited by the Tayebat archivefull text checked★ key study A-INS-13 in-depth review
Second drug after metformin: insulin or sulfonylurea
Association between intensification of metformin treatment with insulin vs sulfonylureas and cardiovascular events and all-cause mortality among patients with diabetes.
Overview
Roumie CL et al.. JAMA 2014 · PMID 24915260 · DOI 10.1001/jama.2014.4312
Veterans who added insulin died more often than those who added a sulfonylurea, but heart attacks and strokes were not more frequent.
Does the study support the specific claim? Partly · Methodological quality: moderate
Plain-language summary
Study explained simply
- What did the researchers want to find out?
Adding insulin vs sulfonylurea to metformin: AMI, stroke, death
- How was the study done?
Retrospective cohort with propensity matching and marginal structural models (VHA, Medicare, NDI)
- How many people, animals or samples were studied?
178,341 (matched 2,436 insulin vs 12,180 SU); Veterans initially treated with metformin
- What was actually found?
Composite aHR 1.30; death aHR 1.44; CV events not increased.
- How large was the observed effect?
aHR 1.30 (1.07-1.58)
- What do the results mean in everyday life?
Possibly reflects sicker patients getting insulin.
- What are the study's limitations?
Residual confounding (sicker patients get insulin); median follow-up about 1 year; mostly male veterans; extra deaths were mainly from cancer and other non-heart causes.
- What does it say about Elawady's specific claim?
Association for C-INS-05; mixed for C-INS-03 (no CV increase).
- Do the results apply to Germany and Europe?
US veterans.
- What can explicitly not be concluded from this study?
That insulin causes death.
Scientific in-depth review
1. Full reference
- Authors
- Roumie CL, Greevy RA, Grijalva CG, Hung AM, Liu X, Murff HJ, Elasy TA, Griffin MR
- Journal
- JAMA
- Year
- 2014
- Identifier
- PMID 24915260 DOI 10.1001/jama.2014.4312 · PMC4149288
- Verification
- verified – PubMed E-utilities efetch, 2026-10-25
- Retracted
- no
- Corrections
- none
- Extent checked
- full text
- Conflicts of interest and funding
Funded by the US Agency for Healthcare Research and Quality (DEcIDE 2 contract), VA CSR&D grant and NIH/NIDDK P30DK092986; AHRQ and VA reviewed the manuscript but had no other role. Authors report no disclosures.
2. Research question
Adding insulin vs sulfonylurea to metformin: AMI, stroke, death
3. Study design and population
- Study design
- Cohort study – Retrospective new-user cohort (Veterans Health Administration, Medicare/Medicaid, National Death Index) of metformin initiators who intensified with insulin or a sulfonylurea; 1:5 propensity-score matching plus marginal structural Cox models with stabilised, truncated IPTW; on-treatment (persistence-required) primary analysis.
- Population
US veterans (95% male, ~70% white), median age 60, incident metformin users 2001-2008 with regular VHA care; median 14 months on metformin and median HbA1c 8.1% at intensification; hospice/dialysis excluded.
- Country
USA
- Period
Metformin initiation October 2001 - September 2008; follow-up to September 2011 (cause of death to September 2009); median time to censoring/outcome 1.15 years.
- Sample size
178,341 metformin initiators; 2,948 added insulin and 39,990 a sulfonylurea; matched cohort 2,436 metformin+insulin vs 12,180 metformin+sulfonylurea (14,616).
- Inclusion and exclusion criteria
Adherent to metformin at intensification; >=365 days baseline; no diabetes drug in prior 180 days; follow-up began 180 days after intensification; censoring at 90 days without metformin, addition of a third drug or loss to follow-up. 512 unmatched insulin users (median HbA1c 11.8%, 67% recently hospitalised) excluded from matched analysis.
4. Methods and endpoints
- Exposure / intervention
Metformin + insulin (47% long-acting, 22% long+short, 17% premixed, 11% short-acting).
- Comparator
Metformin + sulfonylurea
- Endpoints
Primary: AMI, stroke hospitalisation or all-cause death; secondary: AMI/stroke; all-cause death; AMI, stroke or CV death; cause-specific deaths.
- Statistical methods
Propensity score (incl. HbA1c, creatinine, BMI, BP, LDL, comorbidities, utilisation) 1:5 optimal matching; MSM with time-varying covariates; multiple imputation; sensitivity analyses (no persistence requirement, weight truncation), subgroups, quantitative bias analysis for unmeasured confounding.
5. Actual results
- Results
Primary composite 172 vs 634 events, 42.7 vs 32.8 per 1,000 person-years, aHR 1.30 (95% CI 1.07-1.58; P=0.009). AMI/stroke 41 vs 229, 10.2 vs 11.9 per 1,000 PY, aHR 0.88 (0.59-1.30). All-cause death 137 vs 444, 33.7 vs 22.7 per 1,000 PY, aHR 1.44 (1.15-1.79). AMI, stroke or CV death 54 vs 258, 22.8 vs 22.5 per 1,000 PY, aHR 0.98 (0.71-1.34). Cause-specific deaths: CV aHR 1.21 (0.74-2.00), cancer 1.85 (1.21-2.84), other 1.36 (0.90-2.04). Without persistence requirement: aHR 1.29 (1.15-1.46). An unmeasured confounder (HR 1.25) would need to be 30% (composite) or 70% (mortality) more prevalent among insulin users to remove significance.
- Absolute risks
Composite 42.7 vs 32.8 per 1,000 person-years (about 9.9 more events per 1,000 PY); all-cause death 33.7 vs 22.7; AMI/stroke 10.2 vs 11.9 per 1,000 PY.
- Authors' conclusion
Adding insulin vs sulfonylurea to metformin was associated with an increased risk of the composite of non-fatal CV outcomes and all-cause mortality; findings require further investigation; residual confounding by frailty or diabetes severity remains possible; significance driven by all-cause mortality.
6. Strengths
- Active-comparator new-user design in patients eligible for either drug
- Rich clinical covariates (HbA1c, creatinine, BMI, BP, LDL) and time-varying adjustment via MSM
- Good balance after matching; quantitative bias analysis; no industry funding or COI
7. Weaknesses
- Residual confounding by frailty/diabetes severity (authors); before matching insulin users had far more hospitalisations (35% vs 14% in last year) and comorbidities
- Excess limited to non-CV, especially cancer deaths (aHR 1.85) within ~1 year - implausible as a drug effect and suggestive of confounding/reverse causation
- Short median follow-up (1.15 years); high censoring (~60% at therapy change) under persistence requirement
- Small insulin group (7% of intensifiers); 95% male veterans
- Cause of death available for only 76% of deaths
8. Risk of bias
- Orienting tool
- ROBINS-E
- Type of assessment
- structured assessment based on the full text by a single reviewer (no formal double review)
- Full formal assessment
- no
- Overall risk of bias
- some concerns
- Rationale
Single-reviewer ROBINS-E-oriented judgement from the full text; formal assessment requires two independent reviewers. Overall unchanged (some_concerns): methodologically the strongest of the archive's insulin cohorts, but the pattern of excess (non-CV/cancer deaths, no CV excess) points to residual confounding.
| Domain | Judgement | Rationale |
|---|---|---|
| Confounding | some concerns | Extensive measured confounder control with good balance, but strong pre-matching imbalance in hospitalisation/comorbidity indicates channelling of sicker patients to insulin; frailty not measured. |
| Exposure measurement | low | Pharmacy fill data with persistence definitions. |
| Selection of participants | some concerns | Matched population excludes the sickest insulin users; follow-up starts 180 days after intensification (excludes early events). |
| Post-exposure interventions | some concerns | Censoring at therapy change is high and potentially informative, partly addressed by PENR sensitivity analysis. |
| Missing data | some concerns | Missing laboratory values (19-30%) multiply imputed; cause of death missing for 24%. |
| Outcome measurement | low | Validated AMI/stroke codes and vital status/NDI. |
| Selection of reported result | low | Pre-specified primary and secondary outcomes reported, incl. null CV results. |
Full-text review (25 October 2026): Design details (matching, MSM, persistence, 180-day lag), insulin types, cause-specific death results (cancer aHR 1.85), PENR sensitivity (1.29), bias analysis and pre-matching imbalances added; follow-up median 1.15 years (not 14 months, which is time on metformin) corrected; funding AHRQ/VA/NIH and no disclosures; ROBINS-E domains added, overall unchanged.
9. Assessment of the statistical interpretation
Composite excess driven entirely by all-cause death; CV events (AMI/stroke, CV death) were not increased (aHR 0.88 and 0.98). An excess of cancer deaths within ~1 year of adding insulin is more plausibly explained by underlying illness than by insulin. Does not support insulin as a driver of atherosclerosis.
10. Relevance to the specific claim
Association for C-INS-05; mixed for C-INS-03 (no CV increase).
11. Independent contrary findings
12. Transferability to Germany and Europe
US veterans.
13. Overall methodological judgement
Well-conducted cohort showing higher all-cause (mainly cancer and other non-CV) mortality after adding insulin rather than a sulfonylurea, with no increase in CV events; residual confounding likely. Not evidence that insulin causes atherosclerosis or doubles mortality.
| Multidimensional assessment | |
|---|---|
| Authenticity of the source | verified |
| Methodological quality | moderate |
| Risk of bias | some concerns |
| Precision of results | moderate |
| Directness | direct |
| Reproducibility | not replicated |
| Consistency with the overall evidence | unknown |
| Regional transferability | moderate |
| Adequacy of the interpretation (by the archive) | essentially correct but incomplete |
14. Open questions
None beyond those stated.
15. Linked fact checks
- Insulin (and injected insulin) is a principal driver of atherosclerosis. – Studies contradicting or limiting the claim
- Insulin therapy in type 2 diabetes roughly doubles all-cause mortality and raises cancer, renal and cardiovascular events, with a dose-response (each extra IU/kg/day +54% mortality), compared with metformin or sulfonylureas. – Studies supporting the claim
Use by the Tayebat archive
Note: according to the archive, the studies were compiled by the archive operators. Whether Dr. Elawady himself cited this study is not established.
- What the archive says about the study
Roumie (JAMA 2014): adding insulin to metformin raised the composite of MI/stroke/death 30% and all-cause mortality 44% vs adding a sulfonylurea.
- A. What was actually studied?
Adding insulin vs sulfonylurea to metformin: AMI, stroke, death
- B. What did the authors find?
Composite aHR 1.30 (1.07-1.58); all-cause death aHR 1.44; AMI/stroke aHR 0.88 and AMI/stroke/CV death aHR 0.98 (not increased); excess concentrated in cancer deaths (aHR 1.85) over median 1.15 years.
- C. What conclusions do the authors themselves draw?
Adding insulin vs SU was associated with increased composite of nonfatal CV outcomes and all-cause mortality; requires further investigation.
- D. What does the archive derive from it?
Used to argue sulfonylurea before insulin.
- E. Is this derivation scientifically justified?
Numbers correct; the archive omits that MI/stroke and CV death were not increased and that the authors themselves flag residual confounding by frailty/severity; the rapid cancer-death excess argues against a causal insulin effect.
- Adequacy of the interpretation (by the archive)
- essentially correct but incomplete