Cohort studycited by the Tayebat archivefull text checked★ key study A-INS-04 in-depth review
German insurance study on glargine and cancer
Risk of malignancies in patients with diabetes treated with human insulin or insulin analogues: a cohort study.
Overview
Hemkens LG et al.. Diabetologia 2009 · PMID 19565214 · DOI 10.1007/s00125-009-1418-4
Without adjustment glargine users actually had slightly fewer cancers; only after statistical dose adjustment did glargine look somewhat worse, and follow-up was under 2 years.
Does the study support the specific claim? Partly · Methodological quality: low
Plain-language summary
Study explained simply
- What did the researchers want to find out?
Cancer and mortality risk with human insulin vs insulin analogues
- How was the study done?
Retrospective cohort from statutory health insurance claims (AOK)
- How many people, animals or samples were studied?
127,031; Patients starting insulin monotherapy without known cancer
- What was actually found?
Dose-adjusted cancer HR up to 1.31 for glargine; no increase for other analogues.
- How large was the observed effect?
Crude incidence 2.14 (glargine) vs 2.50 (human insulin) per 100 patient-years; dose-adjusted HR 1.09-1.31.
- What do the results mean in everyday life?
An early warning signal that later trials did not confirm.
- What are the study's limitations?
Follow-up about 1.6 years; result reverses with dose adjustment; dose computed over time at risk; no data on BMI, smoking, diabetes duration; groups clinically different.
- What does it say about Elawady's specific claim?
Main support for C-INS-06.
- Do the results apply to Germany and Europe?
German data; directly relevant.
- What can explicitly not be concluded from this study?
That glargine causes cancer.
Scientific in-depth review
1. Full reference
- Authors
- Hemkens LG, Grouven U, Bender R, Günster C, Gutschmidt S, Selke GW, Sawicki PT
- Journal
- Diabetologia
- Year
- 2009
- Identifier
- PMID 19565214 DOI 10.1007/s00125-009-1418-4 · PMC2723679
- Verification
- verified – PubMed E-utilities efetch, 2026-10-25
- Retracted
- no
- Corrections
- none
- Extent checked
- full text
- Conflicts of interest and funding
Authors declare no duality of interest; no funding statement in the full text (authors from IQWiG and the AOK research institute WIdO).
2. Research question
Cancer and mortality risk with human insulin vs insulin analogues
3. Study design and population
- Study design
- Cohort study – Retrospective (pragmatic) cohort study using pseudonymous claims data of the AOK statutory health insurance (13 of 16 federal states, 17.9 million insurants); hypothesis-generating, no adjustment for multiple testing.
- Population
Adults (>=18 years) starting first-time insulin therapy exclusively with one insulin type (human insulin, aspart, lispro or glargine), insulin-naive for 1 year and without any cancer diagnosis or suspicion for 3 years before entry; type 1 vs type 2 diabetes not distinguishable (mostly type 2: 77-92% on oral agents). Mean age 66-70 years.
- Country
Germany
- Period
Data January 1998 - June 2005; inclusion from 1 January 2001; follow-up to 30 June 2005.
- Sample size
127,031 (of 322,732 insulin-treated patients): human insulin 95,804; aspart 4,103; lispro 3,269; glargine 23,855. 5,009 malignant neoplasms, 18,253 deaths.
- Inclusion and exclusion criteria
Excluded: any concomitant other insulin, porcine/bovine insulin, prior cancer or suspected cancer (incl. ICD Z03.1), mean daily dose >1000 IU, event on inclusion date; continuous follow-up required (>=1 insulin prescription every 6 months).
4. Methods and endpoints
- Exposure / intervention
Glargine, aspart or lispro monotherapy; dose = cumulative prescribed dose divided by time at risk (mean daily dose).
- Comparator
Human insulin only (could include short- and long-acting human insulins).
- Endpoints
Primary: diagnosis of malignant neoplasm (ICD-10 C00-C97, D00-D09) from hospital claims; secondary: all-cause mortality.
- Statistical methods
Cox proportional hazards with backward selection (alpha 0.01) of covariates (age, sex, dose, oral agents, concomitant medication, hospitalisation, federal state, year) and insulin-by-dose interaction; stratified Cox for non-proportional dose; sensitivity analyses (excluding skin cancer, in-situ lesions, follow-up >=1.5 years, oral-agent users). No adjustment for multiple testing.
5. Actual results
- Results
Mean follow-up 1.63 years (median 1.41, max 4.41); glargine 1.31 years. Crude cancer incidence per 100 patient-years: human insulin 2.50, aspart 2.16, lispro 2.13, glargine 2.14. Unadjusted HR glargine vs human insulin 0.85 (95% CI 0.79-0.93); age/sex-adjusted 0.86 (0.79-0.94); after adding dose 1.14 (1.05-1.24); final model with dose interaction 1.09 (1.00-1.19) at 10 IU/day, 1.19 (1.10-1.30) at 30 IU, 1.31 (1.20-1.42) at 50 IU. Aspart (p=0.30) and lispro (p=0.96) not different. Excluding skin cancer: glargine HR 0.91 (0.83-0.99) at 10 IU, 1.03 (0.94-1.12) at 30 IU, 1.16 (1.06-1.27) at 50 IU. Dose subgroups: glargine vs human insulin HR 1.13 (0.98-1.30) <20 IU, 1.01 (0.88-1.16) 20-40 IU, 1.59 (1.30-1.94) >40 IU (only 3,226 glargine patients, 103 events). Higher dose associated with cancer for all insulins (HR per 1 SD: glargine 1.78, human insulin 1.57). Mortality: final-model glargine HR 0.76 (0.70-0.83) at 10 IU, 1.20 (1.11-1.30) at 50 IU.
- Absolute risks
Crude cancer incidence per 100 patient-years: human insulin 2.50, aspart 2.16, lispro 2.13, glargine 2.14; in >40 IU dose group glargine 5.26 (4.29-6.38) vs human insulin 3.10 (2.96-3.23).
- Authors' conclusion
Considering the dose-cancer relationship and the lower glargine dose, cancer incidence with glargine was higher than expected; observational data support safety concerns about glargine's mitogenic properties; a non-randomised study cannot prove causality; prospective randomised long-term studies are needed.
6. Strengths
- Large, population-based claims cohort representative of insured insulin users in Germany
- Exclusion of prevalent/suspected cancer at baseline; new-user, single-insulin design
- Transparent stepwise model reporting (Tables 2-3) and multiple sensitivity analyses
- No declared conflicts of interest
7. Weaknesses
- Very short follow-up (mean 1.63 years; glargine 1.31 years), implausibly short for insulin-induced carcinogenesis; detection or promotion of pre-existing tumours cannot be separated
- Direction of the glargine effect reverses on dose adjustment (crude HR 0.85 -> adjusted 1.09-1.31); result depends on the dose-interaction model
- Dose defined as cumulative dose divided by time at risk, i.e. a post-baseline, follow-up-dependent variable (risk of time-related bias); dose also a marker of insulin resistance/obesity
- No data on BMI, smoking, diabetes duration, diabetes type, HbA1c, socioeconomic status
- Treatment groups clinically dissimilar (glargine mostly basal add-on to oral agents vs human insulin often full regimens) - confounding by indication/channelling
- Effect concentrated in the high-dose subgroup with few glargine patients; partly driven by skin cancer at low doses; many models without multiplicity correction
- Paradoxical mortality pattern at high glargine doses unexplained
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
- high
- Rationale
Single-reviewer structured ROBINS-E-oriented judgement from the full text; a formal assessment requires two independent reviewers. Overall high risk of bias unchanged versus abstract-based rating; full text adds the crude/unadjusted inverse association, the time-at-risk-based dose definition and the concentration of the signal in >40 IU users.
| Domain | Judgement | Rationale |
|---|---|---|
| Confounding | high | BMI, smoking, diabetes duration/type, glycaemic control unavailable; glargine and human insulin users clinically different (oral agents 92% vs 77%, fewer hospital stays). |
| Exposure measurement | high | Exposure from prescriptions; mean daily dose computed over time at risk, which depends on time to event; dose-by-insulin interaction drives the result. |
| Selection of participants | some concerns | Restriction to single-insulin users excludes patients switching or combining insulins; requires continuous prescriptions (survivor/adherence selection). |
| Post-exposure interventions | some concerns | Patients who added another insulin were excluded rather than analysed; treatment intensification not modelled. |
| Missing data | some concerns | Bavaria data incomplete; three federal states missing; no information on loss to follow-up beyond insurance cancellation. |
| Outcome measurement | some concerns | Cancer from hospital claims diagnoses including in-situ/skin lesions; detection bias possible (more hospitalisation in human-insulin group); sensitivity analyses partly address this. |
| Selection of reported result | some concerns | Many intermediate models and dose levels presented without multiplicity correction; final model chosen by data-driven backward selection including interaction terms. |
Full-text review (25 October 2026): Group sizes, events, crude incidence rates (2.14 vs 2.50/100 PY) and unadjusted HR 0.85 added; stepwise models, skin-cancer and dose-subgroup sensitivity results added (signal concentrated in >40 IU); dose definition (cumulative dose/time at risk) and exclusions documented; COI: no duality of interest; ROBINS-E domains added, overall remains high; archive-use fields updated.
9. Assessment of the statistical interpretation
The positive glargine signal emerges only after dose adjustment with a dose-by-insulin interaction; crude and age/sex-adjusted HRs were below 1. With mean follow-up of 1.3-1.6 years and a time-at-risk-based dose variable, the association is more consistent with confounding/time-related bias or detection than with carcinogenesis. Absolute crude incidence was lower with glargine (2.14 vs 2.50 per 100 patient-years).
10. Relevance to the specific claim
Main support for C-INS-06.
11. Independent contrary findings
A-INS-07, I-INS-05, I-INS-06, A-INS-05 (overall null).
12. Transferability to Germany and Europe
German data; directly relevant.
13. Overall methodological judgement
Hypothesis-generating signal; not confirmed by RCT evidence.
| Multidimensional assessment | |
|---|---|
| Authenticity of the source | verified |
| Methodological quality | low |
| Risk of bias | high |
| Precision of results | moderate |
| Directness | direct |
| Reproducibility | not replicated |
| Consistency with the overall evidence | inconsistent |
| Regional transferability | high |
| Adequacy of the interpretation (by the archive) | essentially correct but incomplete |
14. Open questions
Very long-term breast cancer risk.
15. Linked fact checks
- Insulin glargine increases cancer risk in a dose-dependent way (and long-term use increases breast cancer). – Studies with mixed or unclear results
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
Hemkens (Diabetologia 2009): in 127,031 German patients, insulin glargine showed a dose-dependent rise in cancer incidence vs human insulin (HR 1.09/1.19/1.31 for 10/30/50 IU/day).
- A. What was actually studied?
Cancer and mortality risk with human insulin vs insulin analogues
- B. What did the authors find?
Crude cancer incidence lower with glargine (2.14 vs 2.50 per 100 patient-years; unadjusted HR 0.85); after dose adjustment HR 1.09/1.19/1.31 at 10/30/50 IU/day; signal concentrated in >40 IU users; no increase for aspart or lispro; mean follow-up 1.63 years.
- C. What conclusions do the authors themselves draw?
Observational data support safety concerns; a non-randomised study cannot prove causality; prospective randomised long-term studies are needed.
- D. What does the archive derive from it?
Used to support that insulin (glargine) raises cancer risk.
- E. Is this derivation scientifically justified?
Numbers are quoted correctly, but the derivation is not justified as evidence of causation: the archive omits that crude incidence was lower with glargine, that the excess appears only after modelling dose (computed over a 1.3-1.6-year follow-up), that the authors explicitly state causality cannot be proven, and that the randomised ORIGIN trial (also in the archive's bibliography) found no cancer increase.
- Adequacy of the interpretation (by the archive)
- essentially correct but incomplete