BGPT: Paper Review: Comparative effectiveness of sitagliptin vs sulphonylureas in older people

Explore by Goal

Quick Explanation Copied

Core finding (real-world, UK primary care, THIN): In adults with T2DM on metformin add-on therapy, sitagliptin produced similar HbA1c reductions versus sulphonylureas in both older (≥75) and younger (18–75) groups, while sitagliptin was associated with greater weight loss and there was a signal of possible over-treatment with sulphonylureas in older adults (more people recorded HbA1c < 48 mmol/mol).

Long Explanation

Paper review: Comparative effectiveness of sitagliptin vs sulphonylureas in older people

DOI: 10.1093/ageing/afz078 | Study type: retrospective cohort (THIN, UK primary care)

1) What the study claims

Design/population: Retrospective cohort using THIN (UK primary care), comparing sitagliptin vs sulphonylureas as add-on to metformin with outcomes measured at ~12 months; age-stratified into older (≥75) and younger (18–75).
HbA1c (older): Adjusted mean HbA1c difference at 12 months was +1.0 mmol/mol (95% CI -0.7 to 2.8), i.e., not statistically significant.
Weight (older): Adjusted mean weight difference at 12 months was -1.4 kg (95% CI -2.1 to -0.7) favoring sitagliptin.
Over-treatment signal (older): Odds of recording HbA1c < 48 mmol/mol by study end were lower with sitagliptin (OR 0.63, 95% CI 0.42–0.95).

Skeptical note: This is effectiveness in observational practice data, so causal interpretation is limited by residual confounding and by how HbA1c/targets and follow-up selection work in routine care.

2) Visualizing the reported effect sizes

Forest-style plots below summarize adjusted mean differences (sitagliptin − sulphonylurea) and odds ratios for HbA1c target achievement as reported.

Values come from the reported adjusted estimates: MD +1.0 mmol/mol (95% CI -0.7 to 2.8).

Values come from the reported adjusted weight estimate: MD -1.4 kg (95% CI -2.1 to -0.7).

ORs and CIs are taken directly from the paper: older OR 0.63 (95% CI 0.42–0.95); younger OR 0.75 (95% CI 0.66–0.86).

3) Method quality & validity checks (what could break the inference)

3.1 Confounding & channeling

Because treatment choice is not randomized, remaining confounding is plausible even after multivariable adjustment. The paper states it adjusts for baseline differences and a priori-selected covariates, but observational residual confounding can remain.

3.2 Endpoint construction and measurement

HbA1c thresholds used as proxies for “over-treatment” may not directly equal hypoglycaemia risk because hypoglycaemia depends on factors beyond HbA1c level (e.g., drug exposure, renal function dynamics, meal patterns). The paper explicitly limits its own safety evaluation and interprets HbA1c <48 as potentially unnecessary/over-treatment in older adults, referencing hypoglycaemia concern.

3.3 Follow-up selection / missing data

The paper states it investigates differences among those with and without missing data for relevant covariates and follow-up duration. In observational settings, missingness can still bias results if missingness is related to both treatment and outcome in ways not fully captured by measured covariates.
The methods reference general epidemiological guidance on missing outcome data handling (to which readers can compare whether the study used appropriate strategies).

3.4 Adherence surrogate

The paper uses a prescription-gap-based “adherent” subgroup and notes this is a surrogate, not true adherence—so residual bias is still possible if prescription patterns differ systematically by patient frailty, care intensity, or clinician preference.

4) Biological plausibility & external consistency (with skepticism)

4.1 HbA1c similarity is plausible

The study’s “similar HbA1c reductions” in older and younger groups can be cross-checked against prior comparative evidence in elderly populations, though effect sizes across studies vary by design, eligibility criteria, and background therapy.
One elderly-focused trial comparison reported broadly comparable glycaemic effects of sitagliptin and a sulphonylurea in older patients.

4.2 Weight difference direction is also plausible

GLP-1/DPP-4 class mechanisms often differ from sulphonylureas in effects on appetite and weight; however, within this paper, the weight difference is empirical and specific (MD -1.4 kg older, MD -2.3 kg younger).

4.3 “Over-treatment” signal: HbA1c <48 as a proxy for hypoglycaemia risk (not direct evidence)

The paper connects lower HbA1c targets—especially <48 mmol/mol—to concern about unnecessary tight control in older adults and associated hypoglycaemia risk with sulphonylureas.
General evidence shows sulphonylureas carry higher hypoglycaemia risk than DPP-4 inhibitors in clinical comparisons, and this risk is often amplified in older adults.
Broader comparative reviews also describe differences between DPP-4 inhibitors and sulphonylureas, including hypoglycaemia risk profiles.

Disproof-oriented framing: This interpretation would be weakened if (a) HbA1c <48 does not correlate with hypoglycaemia events in similar observational cohorts after adjusting for frailty/renal function, or (b) sulphonylurea users differ systematically in the propensity to measure HbA1c at the right timepoints.

5) Statistical results distilled into tables

Outcome	Age group	Effect (sitagliptin − sulphonylurea)	Uncertainty	Direction
HbA1c at 12 months (adjusted mean difference)	Older (≥75)	+1.0 mmol/mol	95% CI: -0.7 to +2.8	No clear advantage (non-significant)
Weight at 12 months (adjusted mean difference)	Older (≥75)	-1.4 kg	95% CI: -2.1 to -0.7	Favors sitagliptin (lower weight)
HbA1c <48 mmol/mol by end (adjusted OR)	Older (≥75)	0.63	95% CI: 0.42 to 0.95	Favors sitagliptin (fewer <48 recordings)
HbA1c at 12 months (adjusted mean difference)	Younger (18–75)	+0.8 mmol/mol	95% CI: 0.2 to 1.4	Small difference (authors: not clinically significant)
Weight at 12 months (adjusted mean difference)	Younger (18–75)	-2.3 kg	95% CI: -2.5 to -2.0	Favors sitagliptin (lower weight)
HbA1c <48 mmol/mol by end (adjusted OR)	Younger (18–75)	0.75	95% CI: 0.66 to 0.86	Favors sitagliptin (fewer <48 recordings)

All numbers above are directly from the provided paper text.

6) Limitations & blind spots (what we should not over-interpret)

Observational design: residual confounding remains possible despite multivariable adjustment; treatment selection may correlate with unmeasured clinical factors (frailty, clinician preferences, renal trajectories).
Safety outcome scope: the report emphasizes effectiveness and interprets HbA1c target signals as “over-treatment” concern; it is not presented as a direct hypoglycaemia event study within the extracted text.
Drug heterogeneity within classes: UK prescribing may differ within the sulphonylurea and DPP-4 inhibitor subtypes; the paper focuses on sitagliptin vs “sulphonylureas” broadly and discusses power limitations for other DPP-4 inhibitors.
Generalizability: results are anchored to UK primary care practices and THIN capture patterns during 2007–2014, and newer agents are now common.

What would disprove/meaningfully change the interpretation?

If future analyses in similar datasets with improved hypoglycaemia outcome capture (and stronger causal methods) show no difference in clinically meaningful hypoglycaemia events despite HbA1c<48 target differences.
If randomized evidence in older adults shows clinically meaningful HbA1c or weight superiority for one class, contrary to the observed lack of HbA1c advantage. (The paper’s claim is about comparative effectiveness, not definitive causal superiority.)

Author-specific follow-ups

Feedback:

Updated: April 13, 2026