Open questions — type 2 diabetes¶
Last curated: 2026-10-30
Stable IDs persist across future updates. A question moves to “answered” only when the relevant outcome—not merely a surrogate—has adequate evidence.
Tier 1 — practice-changing and designable now¶
OQ-1 — How should remission durability be increased?¶
DiRECT remission fell from 46% at one year to 36% at two years and 13% in the five-year extension group; maintained weight loss is strongly associated with persistence, but β-cell reserve and environment may explain additional relapse (Lean 2018, PMID 29221645; Lean 2019, PMID 30852132; Lean 2024, PMID 38423026). A factorial maintenance trial could compare behavioural intensity, intermittent formula diet and pharmacologic maintenance, stratified by duration/C-peptide.
OQ-2 — What should drug-dependent normoglycaemia be called and measured as?¶
Consensus remission requires HbA1c <6.5% without glucose-lowering medication, excluding clinically valuable normoglycaemia during GLP-1/tirzepatide therapy (Riddle 2021, PMID 34462270; Frías 2021, PMID 34170647). A 2026 review now proposes “pharmacological remission,” and a 14,141-person observational study produced frequencies from 5.8% to 18.3% under four definitions (Bianchi 2026, PMID 41852334; Fadini 2025, PMID 41142657). The unresolved task is consensus validation against off-treatment durability, complications and treatment burden.
OQ-3 — Is routine GLP-1RA plus SGLT2 therapy superior to either alone on hard outcomes?¶
Indirect evidence suggests complementary strengths—SGLT2 for HF, GLP-1RA for stroke and both for kidney/mortality. GLP-1RA effects appear consistent with baseline SGLT2 use, but only 1,743/17,072 participants in the relevant meta-analysis used SGLT2 at baseline; this tests interaction, not randomised incremental combination benefit (Palmer 2021, PMID 33441402; Neuen 2024, PMID 39210781). A 2026 target-trial emulation associated triple RAASi+SGLT2i+GLP-1RA therapy with lower mortality and kidney-event risk, but residual confounding remains (Casper 2026, PMID 42658186). The 2026 ADA/EASD consensus now considers earlier combination use in CVD, CKD and HF, but its recommendation does not settle the incremental randomised hard-outcome comparison (Davies 2026, PMID 42825450). A pragmatic factorial outcome trial in dual cardiorenal risk is feasible.
OQ-4 — Does early organ-protective therapy create a legacy effect?¶
UKPDS demonstrates benefits persisting 24 years after early glycaemic assignment, but whether finite early SGLT2/GLP-1 exposure changes long-term trajectories is unknown (Adler 2024, PMID 38772405). Withdrawal or delayed-start extensions with ethical rescue rules could test persistence. A 2026 modified-Delphi core outcome set for disease modification (35 outcomes across 11 domains; round one n=233, round two n=79) now gives such trials a common outcome list to report against (Khunti 2026, PMID 42810354).
OQ-5 — Which modern strategy is best over ten years: surgery, incretin therapy or structured diet?¶
Each approach achieves substantial glycaemic/weight benefit in unlike trials; surgery has longer follow-up, incretins have large CVOTs, and DiRECT has primary-care remission evidence (Courcoulas 2024, PMID 38411644; Aronne 2025, PMID 40353578; Lean 2024, PMID 38423026). A live 2026 search found a systematic review of six retrospective surgery-versus-GLP-1RA cohorts and a direct observational comparison, but no decade-long randomised hard-outcome comparison in the PubMed search (Campbell 2026, PMID 42660781; Gasoyan 2025, PMID 40957960). A preference-aware pragmatic comparative-effectiveness trial should measure hard outcomes, treatment burden and total cost.
OQ-6 — Can youth-onset T2D complication trajectories be altered?¶
By mean age 26.4 years, 60.1% of TODAY follow-up participants had at least one microvascular complication. RISE showed neither glargine→metformin nor metformin alone halted β-cell decline, while DINAMO and SURPASS-PEDS established short-term empagliflozin and tirzepatide glycaemic efficacy without organ-outcome evidence in youth (TODAY 2021, PMID 34320286; RISE 2018, PMID 29941500; Laffel 2023, PMID 36738751; Hannon 2025, PMID 40975112; NCT05260021). A live search through 2026-08-30 found efficacy and cohort evidence but no paediatric organ-outcome strategy trial; long trials need β-cell, kidney, retina and cardiovascular endpoints rather than adult extrapolation.
OQ-7 — How can SGLT2 sick-day rules reduce DKA without avoidable permanent discontinuation?¶
Euglycaemic DKA is rare but diagnostically hazardous, particularly with surgery, fasting and acute illness (Dutta 2022, PMID 35495849). Scandinavian registers sharpen both halves of the question. Ketoacidosis occurred at 2.43 per 1,000 person-years and did not stop being a risk after the early treatment period. Infection co-occurred in 31.8% of cases, the setting in which sick-day rules apply. Only 25.1% were taking an SGLT2 inhibitor one year after an event (Kadesjö 2026, PMID 42679839). The register cannot tell whether that discontinuation was avoidable. A live search through 2026-08-30 found perioperative guidance and outcome syntheses but no randomised sick-day implementation pathway; cluster-randomised studies could compare written rules, EHR prompts, ketone access and restart pathways.
OQ-8 — Does anti-stigma care improve clinical outcomes?¶
Consensus evidence links stigma to distress and care barriers (Speight 2024, PMID 38128969). A 21-clinician randomised pilot improved attitudes toward stigma-free language but did not measure sustained clinician behaviour or any patient outcome (Joiner 2026, PMID 42579884). A larger clinic-level language/training intervention with validated stigma, attendance, medication-persistence and clinical outcomes is now the relevant next test.
OQ-9 — Which biomarker adds enough value to guide treatment?¶
Clusters and multi-omics predict heterogeneity, but simple clinical variables often match cluster-based treatment-response prediction (Dennis 2019, PMID 31047901; Carrasco-Zanini 2024, PMID 37889320). The TriMaster crossover has compared treatment-allocation models retrospectively within trial data, without prospective score-based assignment (Güdemann 2026, PMID 42312889). The searched PubMed records did not establish a randomised test-use strategy with patient outcomes; future trials must randomise use of the test, not merely validate association.
OQ-10 — How should rapid HbA1c improvement be managed in advanced retinopathy?¶
SUSTAIN-6 detected retinopathy complications during rapid improvement, while ACCORD Eye and its follow-on show lower long-term retinopathy progression after intensive glycaemia (Marso 2016, PMID 27633186; Chew 2010, PMID 20587587; ACCORDION Eye 2016, PMID 27289122). A live search through 2026-08-30 found reviews and observational/legacy evidence but no randomised monitoring-or-titration strategy for early worsening; such trials could test whether monitoring or slower reduction preserves long-term benefit.
Tier 2 — important but requiring method or infrastructure development¶
OQ-11 — Is there a stable mechanistic taxonomy of T2D?¶
Data-driven clusters reproduce phenotypic variation but depend on continuous variables, timing and treatment; genetic pathways demonstrate heterogeneity without a validated clinical taxonomy (Ahlqvist 2018, PMID 29503172; Suzuki 2024, PMID 38374256). A live search through 2026-08-30 found cluster validation and response-prediction studies, but no randomised cluster-guided outcome strategy.
OQ-12 — What is the operational biomarker of the personal fat threshold?¶
The twin-cycle model explains remission through ectopic-fat reduction, yet no scalable liver/pancreas-fat or adipose-capacity threshold guides care across ancestries (Taylor 2024, PMID 39038473).
Coverage limit for cross-region interpretation: SciELO and LILACS could not be searched (HTTP 403 access limitation documented in SOURCES.md). This sweep cannot establish transportability across ancestries from the reachable databases alone.
OQ-13 — How should global incidence be measured consistently?¶
IDF and GBD reconcile unlike tests, age ranges and source quality; low-resource settings have the largest diagnostic gaps (Sun 2022, PMID 34879977; GBD 2021 Diabetes Collaborators 2023, PMID 37356446). A standard repeated biochemical-surveillance platform is needed.
OQ-14 — Which prevention policies work without widening inequity?¶
High BMI and diet account for large modelled burden fractions, but comparative-risk models do not identify locally effective, equitable policies (GBD 2021 Diabetes Collaborators 2023, PMID 37356446; O'Hearn 2023, PMID 37069363). Among 3.73 million privately insured US adults with diagnosed prediabetes, an instrumental-variable analysis estimated 1.94 (95% CI 1.44–2.44) fewer incident diabetes cases per 100 National DPP participants versus lower-intensity lifestyle services at 12 months. Only 245,896 had any lifestyle-treatment claim, and 27,703 of those used the National DPP (Chorniy 2026, PMID 42227951). The effect is real-world but small, short-horizon, and confined to the insured; the reach figures are the equity problem.
OQ-15 — How much benefit is lost through discontinuation, shortage and cost?¶
Efficacy trials provide free drug and structured follow-up; global disease scale and expenditure make persistence/access central (Sun 2022, PMID 34879977; González-González 2021, PMID 34244276). Linked pharmacy, clinical and patient-reported datasets are needed.
OQ-16 — What is the correct complication-surveillance schedule after remission?¶
Consensus recommends continued observation because prior glycaemic exposure persists (Riddle 2021, PMID 34462270; Adler 2024, PMID 38772405). A live search through 2026-08-30 found consensus recommendations and observational risk studies, but no trial comparing surveillance intervals or deintensification after remission.
OQ-17 — Can diabetes-related MASLD fibrosis be screened without overdiagnosis?¶
T2D is a high-risk context, but liver enzymes are insensitive and simple fibrosis scores have age-related false positives. A two-step FIB-4→elastography pathway has now been evaluated for fibrosis classification and later liver events specifically in T2D; whether screening improves outcomes still requires an intervention study (Driessen 2025, PMID 38147315; Chen 2026, PMID 41911049).
OQ-18 — How should patient-defined benefit enter treatment algorithms?¶
Preferences vary across route, adverse effects, weight, hypoglycaemia and cost, yet algorithms usually optimise disease outcomes first (González-González 2021, PMID 34244276; Davies 2022, PMID 36148880). In the QBSafe cluster RCT (11 clinicians, 143 patients, a federally qualified health centre with 61.5% Spanish-speaking patients), patient-chosen agenda-setting cards raised clinician satisfaction but did not significantly change illness intrusiveness (−3.9, 95% CI −10.4 to 2.6), HbA1c, treatment burden or distress at 6 months (Lipska 2026, PMID 42895748). Eliciting priorities alone appears insufficient; the trial was small.
OQ-19 — Which intervention repairs the global diagnosis-to-control cascade?¶
In 2023, an estimated 55.8% of people with diabetes were diagnosed, 91.4% of those diagnosed were treated, but only 41.6% of treated people had optimal glycaemia; in 55 LMIC surveys, only 4.6% received all recommended treatments (Stafford 2025, PMID 40934935; Flood 2021, PMID 35211689). The leak starts after a positive test even in insured high-income settings. Among 27,650 US commercially insured and Medicare adults with laboratory-confirmed diabetes, 66.6% received both a diagnosis code and drug treatment within a year, 19.7% received a code only and 9.6% received neither. Higher comorbidity burden lowered the probability of both (Stimpson 2026, PMID 42832327). Claims cannot capture uncoded lifestyle management. Cluster-randomised health-system trials should compare screening linkage, medicine supply, team care and risk-factor bundles using population-level control—not clinic attendees alone—as denominator.
OQ-20 — Is youth β-cell decline modifiable or only suppressible while treatment continues?¶
RISE found progressive β-cell deterioration on glargine→metformin or metformin, followed by worsening after withdrawal; liraglutide, empagliflozin and tirzepatide improve HbA1c in short youth trials (RISE 2018, PMID 29941500; Hannon 2020, PMID 32985775; Tamborlane 2019, PMID 31034184; Laffel 2023, PMID 36738751; Hannon 2025, PMID 40975112). A live search through 2026-08-30 found no youth T2D treatment-withdrawal study demonstrating durable secretion preservation; a withdrawal-aware trial should pair clamp/C-peptide measures with clinical durability.
OQ-21 — Is remission causal, or a marker of favourable biology and weight response?¶
Look AHEAD remission was associated with CKD HR 0.67 and CVD HR 0.60, and SOS remission with mortality HR 0.71, but remission is a post-randomisation state correlated with weight loss, fitness and β-cell reserve (Gregg 2024, PMID 38233592; Carlsson 2024, PMID 38896851). Mediation analyses and randomised maintenance interventions are needed before surveillance or risk treatment is de-escalated.
OQ-22 — What is the net benefit of different default HbA1c targets in the low-hypoglycaemia era?¶
ACP proposed 7%–8% for most adults while ADA/AACE often support lower individual targets; older intensive-control trials used therapies with different weight and hypoglycaemia profiles from current incretin/SGLT2 strategies (Qaseem 2018, PMID 29507945; Blonde 2022, PMID 35963508). A live search through 2026-08-30 found modern drug-comparison studies but no randomised default-target strategy built around incretin/SGLT2 therapy; such a trial should stratify by age, frailty and drug class.
The 2026 J-DOIT3 landmark analysis adds achieved-HbA1c associations, not a new randomisation of targets. The original multifactorial groups were pooled; sparse events and limited contemporary GLP-1/SGLT2 exposure restrict interpretation. The full text explicitly cautions against assuming safety at very low HbA1c (Noda 2026, PMID 42912779). OQ-22 remains open.
OQ-23 — Can MASLD screening in T2D improve outcomes rather than referrals?¶
Meta-analysis estimates MASLD in 65.33% of people with T2D and advanced fibrosis in 15.49% of biopsy cohorts, but verification bias inflates histologic estimates (Younossi 2024, PMID 38521116). Chen 2026 evaluated sequential FIB-4→elastography for fibrosis classification and liver-event prognosis in T2D, narrowing the diagnostic and prognostic uncertainty. Its cohort design does not test whether offering screening improves outcomes; a randomised screening-to-care pathway remains the question (Chen 2026, PMID 41911049).
OQ-24 — Does CGM improve complications or mainly short-term glycaemia in non-prandial-insulin T2D?¶
MOBILE improved HbA1c by 0.4 points and time-in-range by 15 points over eight months in basal-insulin users; smaller non-insulin studies confound sensor feedback with education (Martens 2021, PMID 34077499; Martens 2025, PMID 39757879). CONNECT has now randomised 283 non-insulin-treated adults in US primary care to CGM versus routine care: HbA1c −0.9 points (95% CI −1.1 to −0.6) and time-in-range +21 points at 26 weeks, industry-funded (Oser 2026, PMID 42878622). This settles short-term glycaemic efficacy in this population more firmly, not the complication question. A live search through 2026-08-30 found CGM measurement, association and short-term randomised studies but no complication-endpoint strategy trial in this population, and a dedicated 2026-10-23 PubMed query for randomised CGM trials with complication or mortality terms (2026-08-28 to 2026-10-23) returned 0 records, a PubMed-only absence; longer pragmatic trials should measure severe hypoglycaemia, complications, distress, equity and total cost.
OQ-25 — How should active-comparator outcome trials alter sequencing?¶
SURPASS-CVOT established tirzepatide noninferiority but not superiority to dulaglutide (MACE HR 0.92, 95.3% CI 0.83–1.01), answering a different question from placebo-controlled SOUL (HR 0.86, 95% CI 0.77–0.96) (Nicholls 2025, PMID 41406444; McGuire 2025, PMID 40162642). ACHIEVE-4 adds a third comparator type. Oral orforglipron was compared against titrated insulin glargine and met a 1.8 noninferiority margin for MACE-4 (HR 0.84, 95% CI 0.59–1.20); severe or clinically significant hypoglycaemia was 6.8% versus 19.2% (Klein 2026, PMID 42815506). It tells sequencing whether an oral agent can replace insulin intensification safely. It does not show cardiovascular benefit against placebo or an injectable GLP-1RA. Decision models need absolute outcomes, weight, adverse effects, discontinuation and cost under contemporary background therapy.
OQ-26 — Which patient-reported outcomes are actionable in routine care?¶
Diabetes distress and self-management are the most consistently endorsed diabetes-specific outcomes, but screening without a response pathway can add burden (Hamilton 2023, PMID 37672919). A small mixed-type randomised trial has tested questionnaire-informed consultations (Kelly 2025, PMID 37350136). The 2026 EASD guideline now supports routine distress assessment and selected interventions, while explicitly calling for trials of complete assessment-to-response pathways (Speight 2026, PMID 42744915). The remaining T2D-specific trial should randomise a defined response pathway and include distress, safety, HbA1c, attendance and clinician workload.
Dots not yet connected¶
| # | Dot A | Dot B | The missing junction | Powers |
|---|---|---|---|---|
| D-1 | In DiRECT, remission (HbA1c below 6.5% after at least 2 months off antidiabetic drugs) was 68/149 (46%) versus 6/149 (4%) at 12 months (OR 19.7, 95% CI 7.8–49.8). Across the whole population it was 6/89 (7%) with 0–5 kg lost and 31/36 (86%) with 15 kg or more lost (Lean 2018, PMID 29221645). At 24 months remission was 53/149 (36%) versus 5/149 (3%) (Lean 2019, PMID 30852132). | Tirzepatide 15 mg lowered HbA1c by 2.30 percentage points at 40 weeks, versus 1.86 with semaglutide 1 mg, in 1,879 adults with baseline HbA1c 8.28% (Frías 2021, PMID 34170647). Consensus remission is HbA1c below 6.5% measured at least 3 months after glucose-lowering drugs stop (Riddle 2021, PMID 34462270). Among 14,141 GLP-1 receptor agonist initiators followed for about 4 years, remission off all diabetes drugs was 5.8% and remission irrespective of ongoing therapy was 18.3%; cardiovascular events were lower only when no new diabetes drug had been started (HR 0.65, 95% CI 0.48–0.88) (Fadini 2025, PMID 41142657). After semaglutide 2.4 mg was stopped in adults with obesity and without diabetes, 11.6 of the 17.3 percentage points lost by week 68 had been regained by week 120 (Wilding 2022, PMID 35441470). | The 2026-10-07 PubMed search did not identify a randomised trial assigning pharmacologic maintenance against continued behavioural support after diet-induced remission. PubMed on 2026-10-07 returned 2 records for that design (query in the 2026-10-07 log). Kempf 2025 (PMID 41113714) pools separate formula-diet trials against separate drug trials. Thom 2021 (PMID 33289256) analyses ghrelin inside DiRECT. Neither assigns a drug after diet-induced remission. The five-year DiRECT extension is not used here: an erratum exists (PMID 38723645) and its correction text was not retrieved. | OQ-1, OQ-2 |
| D-2 | Early sulfonylurea or insulin versus conventional control in UKPDS still showed relative risk reductions up to 24 years after the trial ended: death 10% (95% CI 2–17; absolute 2.7%), myocardial infarction 17% (6–26; absolute 3.3%), microvascular disease 26% (14–36; absolute 3.5%) (Adler 2024, PMID 38772405). | In EMPA-REG, while treatment assignment continued (n=7,020, median observation 3.1 years), empagliflozin versus placebo gave major cardiovascular events in 10.5% versus 12.1% (HR 0.86, 95.02% CI 0.74–0.99), cardiovascular death in 3.7% versus 5.9% (38% relative risk reduction), and heart-failure hospitalisation in 2.7% versus 4.1% (35% relative risk reduction) (Zinman 2015, PMID 26378978). | The 2026-10-07 PubMed search did not identify a randomised withdrawal or delayed-start extension testing whether a kidney, heart-failure, or mortality benefit persists after an SGLT2 inhibitor or GLP-1 receptor agonist is stopped. PubMed on 2026-10-07 returned 20 records (query in the log). The one legacy analysis is observational: mean HbA1c above 8% for 3 years versus at or below 7% carried a later cardiovascular HR of 1.33 (95% CI 1.063–1.365), and starting an SGLT2 inhibitor in the first 1 or 2 years removed that association (interaction p=0.006 and 0.003) (Ceriello 2023, PMID 37547276). The other 19 titles are reviews, case reports, protocols, post-hoc biomarker papers, or cost studies. | OQ-4 |
| D-3 | In FLOW, semaglutide 1 mg versus placebo (n=3,533, median 3.4 years) reduced the kidney primary outcome, 331 versus 410 events (HR 0.76, 95% CI 0.66–0.88) (Perkovic 2024, PMID 38785209). | In CREDENCE, canagliflozin versus placebo (n=4,401, median 2.62 years) reduced the primary outcome, 43.2 versus 61.2 per 1,000 patient-years (HR 0.70, 95% CI 0.59–0.82) (Perkovic 2019, PMID 30990260). In three GLP-1 outcome trials only 1,743/17,072 participants (10.2%) were on an SGLT2 inhibitor at baseline; the kidney composite risk ratio was 0.79 (95% CI 0.66–0.95) and did not differ by that background (P-heterogeneity 0.53) (Neuen 2024, PMID 39210781). A TriNetX emulation (n=19,139, eGFR 20–60) associated RAASi plus SGLT2 inhibitor plus GLP-1 receptor agonist with mortality aHR 0.317 (95% CI 0.234–0.429) and major adverse kidney events aHR 0.509 (0.374–0.693) versus RAASi alone (Casper 2026, PMID 42658186). | The 2026-10-07 PubMed search did not identify a trial randomising GLP-1 receptor agonist plus SGLT2 inhibitor against either class alone on kidney failure or death. PubMed on 2026-10-07 returned 12 randomised records linking both classes, combination, and a kidney measure (query in the log). PRECIDENTD feasibility (n=173, median follow-up 10 months) measured fills and discontinuation: dual therapy filled 53% versus 84% for monotherapy at 4 months, and 49% versus 22% of fillers discontinued, mostly for side effects (Wexler 2026, PMID 41456635). Long 2026 randomised 120 people with early diabetic kidney disease (30 per arm) for 24 weeks; canagliflozin plus semaglutide reduced the albumin-creatinine ratio more than either drug or control (P below 0.05), with no numeric difference reported and no kidney-failure or mortality endpoint (PMID 42170981). The other ten titles are imaging or biomarker analyses, a baseline-use subgroup, or dapagliflozin versus liraglutide. | OQ-3 |
| D-4 | At the end of TODAY follow-up, mean age was 26.4 years and mean diabetes duration 13.3 years (n=500). At least one complication had occurred in 60.1%, and kidney disease in 54.8% (TODAY Study Group 2021, PMID 34320286). This paper counts complications. It does not measure β-cell function. | Adult outcome trials of a few years show organ benefit on treatment: FLOW kidney HR 0.76 (95% CI 0.66–0.88) over a median 3.4 years (Perkovic 2024, PMID 38785209), and CREDENCE primary HR 0.70 (95% CI 0.59–0.82) over a median 2.62 years (Perkovic 2019, PMID 30990260). | The 2026-10-07 PubMed search did not identify a randomised trial in adolescents or youth with type 2 diabetes reporting retinopathy, kidney failure, or major cardiovascular events on a GLP-1 receptor agonist or SGLT2 inhibitor. PubMed on 2026-10-07 returned 0 records (query in the log). Cochrane CENTRAL and SciELO were not searched. | OQ-6 |
| D-5 | In SUSTAIN-6 (n=3,297, 104 weeks), semaglutide versus placebo reduced major cardiovascular events from 8.9% to 6.6% (HR 0.74, 95% CI 0.58–0.95) and increased retinopathy complications (HR 1.76, 95% CI 1.11–2.78) (Marso 2016, PMID 27633186). | Tirzepatide 15 mg lowered HbA1c by 2.30 percentage points at 40 weeks (Frías 2021, PMID 34170647). Intensive glycaemia in ACCORD Eye reduced 4-year retinopathy progression from 10.4% to 7.3% (adjusted OR 0.67, 95% CI 0.51–0.87) (ACCORD Study Group 2010, PMID 20587587). Four years after closeout the rates were 5.8% versus 12.7% (adjusted OR 0.42, 95% CI 0.28–0.63; n=1,310) (ACCORDION Eye Study Group 2016, PMID 27289122). | The 2026-10-07 PubMed search did not identify a trial randomising the speed of a modern HbA1c fall, or the eye monitoring around it, according to baseline retinopathy. PubMed on 2026-10-07 returned 4 records (query in the log): a systematic review (PMID 42541561), intensive versus conventional control in newly diagnosed diabetes (PMID 34234400), a SUSTAIN-6 post-hoc on HbA1c fall and retinopathy (PMID 29178519), and glargine versus NPH early progression (PMID 17479440). Those four were classified from titles; their abstracts were not re-read. A Cochrane Database of Systematic Reviews query on the same topic returned 0. | OQ-10 |
| D-6 | A GWAS of 2,535,601 people (39.7% not of European ancestry), including 428,452 type 2 diabetes cases, mapped 1,289 signals at 611 loci into eight non-overlapping clusters. Partitioned polygenic scores in a further 279,552 people were associated with coronary disease, peripheral artery disease, and end-stage diabetic nephropathy. The abstract gives no hazard ratios (Suzuki 2024, PMID 38374256). | GRADE did not assign drugs by genotype. In 5,047 people on metformin followed for a mean of 5.0 years, confirmed HbA1c of 7.0% or higher occurred at 26.5 and 26.1 per 100 participant-years with glargine and liraglutide, versus 30.4 with glimepiride and 38.1 with sitagliptin. Severe hypoglycaemia occurred in 2.2% on glimepiride and 1.0% on liraglutide (GRADE Study Research Group 2022, PMID 36129996). | In HNF-1α diabetes, a randomised crossover found a 5.2-fold greater fasting-glucose response to gliclazide than to metformin (4.7 versus 0.9 mmol/L) (Pearson 2003, PMID 14575972). That is monogenic diabetes. PubMed on 2026-10-07 returned 12 records for genotype-guided or pharmacogenetic randomised language in type 2 diabetes (query in the log). Pearson is the only randomised drug-selection trial among them. The other titles are association studies, reviews, and a non-randomised antiplatelet cohort (PMID 38106133). A Cochrane Database of Systematic Reviews query returned 0. This PubMed search did not identify a trial assigning metformin, a sulfonylurea, insulin, or an incretin from the common-variant clusters in Suzuki 2024. | OQ-9, OQ-11 |
| D-7 | In a DiRECT mechanistic subgroup, liver and pancreatic fat fell after weight loss and stayed low through 24 months in people maintaining remission; first-phase insulin response remained improved only in those maintaining remission (Taylor 2019, PMID 31866441). This subgroup does not establish an ancestry-specific fat threshold. | In an exploratory secondary analysis of a low-energy meal-replacement study, 15 White Europeans and 12 South Asians with type 2 diabetes were assigned the diet; all 15 and 8/12, respectively, completed it. Weight loss and liver-fat reduction were similar, but the reported reduction in insulin resistance was smaller in South Asians (HOMA-IR −5.7, 95% CI −7.3 to −4.2, versus −8.6, −9.7 to −7.6), as was the reduction in visceral adiposity (−0.43% versus −0.80%) (Athithan 2023, PMID 37811525). | A shared liver-fat response did not imply an identical metabolic response in this small two-group analysis. It neither measures a personal fat threshold nor establishes how to apply one across ancestries. | OQ-12 |
| D-8 | An international consensus (51 members, 18 countries) states that, on average, four in five adults with diabetes experience diabetes stigma and one in five experience discrimination (Speight 2024, PMID 38128969). | A pilot randomised 21 primary-care clinicians to a stigma-language module (n=12) or an active control (n=9). Attitude scores rose by 1.19 (SD 0.68) versus 0.21 (SD 0.44); condition by time F(1,19)=14.27, p=0.001, partial eta-squared 0.43, Cohen d=1.67. Intention change was not significant (d=0.67, p=0.14). No attendance, adherence, or HbA1c outcome was measured (Joiner 2026, PMID 42579884). | The 2026-10-07 PubMed search did not identify a cluster-randomised trial testing whether clinician language training changes adherence, attendance, or HbA1c. The tight PubMed query on 2026-10-07 returned 2 records (query in the log). Salazar 2023 is a medication-adherence study in rural Indian primary care among people with common mental disorders (PMID 38545530, title). Hansen 2012 is a help-seeking study in low-income Latino adults with diabetes and depression (PMID 22367667, title). Neither trains clinician language and then measures a patient clinical outcome. | OQ-8 |
| D-9 | In a protocolised 40-week trial, tirzepatide 5, 10, and 15 mg changed HbA1c by −2.01, −2.24, and −2.30 percentage points. Glucose below 54 mg/dL occurred in 0.6%, 0.2%, and 1.7% on tirzepatide and in 0.4% on semaglutide 1 mg (Frías 2021, PMID 34170647). | A review through May 2020 found 17 studies and 7,296 patients, and no eligible SGLT2-inhibitor preference study. Five studies (2,662 patients) compared a GLP-1 receptor agonist with another glucose-lowering drug; 12 (4,634) compared GLP-1 agents or devices. Once-weekly, simpler regimens were preferred. Certainty was very low, and no single preference percentage is reported (González-González 2021, PMID 34244276). | The 2026-10-07 PubMed search did not identify a trial comparing preference-concordant prescribing with prescribing aimed at the largest HbA1c effect, on persistence or clinical outcomes, for an SGLT2 inhibitor or a GLP-1 or GIP/GLP-1 agent. PubMed on 2026-10-07 returned 0 records (query in the log). | OQ-15, OQ-18 |
| D-10 | ARMMS-T2D followed 262/305 participants (86%) from four US randomised trials for a median of 11 years; 25% of the medical/lifestyle arm later had surgery. The between-group HbA1c difference was −1.4 percentage points (95% CI −1.8 to −1.0) at 7 years and −1.1 (−1.7 to −0.5) at 12 years. Remission was 18.2% versus 6.2% at 7 years and 12.7% versus 0.0% at 12 years. Major cardiovascular events did not differ. The comparator was medical/lifestyle care (Courcoulas 2024, PMID 38411644). | In adults with obesity and without diabetes (n=751), tirzepatide reduced weight by 20.2% (95% CI −21.4 to −19.1) versus 13.7% (−14.9 to −12.6) with semaglutide at 72 weeks (Aronne 2025, PMID 40353578). Six retrospective cohorts (n=208,751) associated metabolic surgery with fewer major adverse cardiovascular events than GLP-1 receptor agonist therapy (RR 0.59, 95% CI 0.46–0.77); all-cause mortality RR was 0.66 (0.41–1.07). The printed final-A1c interval is internally inconsistent and is not quoted (Campbell 2026, PMID 42660781). | The 2026-10-07 PubMed search did not identify a decade-long randomised comparison of an initial metabolic-surgery strategy versus incretin therapy reporting hard outcomes. A PubMed query restricted to randomised head-to-head titles returned 2 records on 2026-10-07 (query in the log). Nigro 2025 randomised 43 people with class I–II obesity to endoscopic sleeve gastroplasty (n=23) or liraglutide (n=20); excess-weight and total-weight loss differed at 3 and 6 months (P=0.001) and no longer differed at 12 months, and comorbidity resolution did not differ at 3, 6, or 12 months (PMID 40779281). Stanley 2026 randomised semaglutide 2.4 mg against placebo after suboptimal weight loss following bariatric surgery (n=70); week-68 weight change was −18.0% versus +0.4%, adjusted difference −19.18 percentage points (95% CI −23.4 to −14.8) (PMID 42174253). That trial starts after surgery. | OQ-5 |
| D-11 | A comparative-risk model attributed 14.1 million (95% uncertainty interval 13.8–14.4 million), or 70.3% (68.8–71.8), of incident type 2 diabetes cases worldwide in 2018 to 11 dietary factors (O'Hearn 2023, PMID 37069363). This is modelled attribution, not an intervention effect. | In a US produce-prescription study of people with diabetes at risk of food insecurity, 2,177 of 9,644 contacted people enrolled (22.6%); 1,450 were assigned the produce benefit and 1,148 activated their cards. This is an observational analysis of reach within a pragmatic randomised trial, not a measure of diabetes prevention (Rader 2026, PMID 42036037). | A large estimated dietary contribution to incident diabetes does not establish who can be reached by a food intervention. The studies use different populations and outcomes; programme reach and equitable uptake must be evaluated locally. | OQ-14, OQ-19 |
| D-12 | The remission consensus defines remission as HbA1c below 6.5% (48 mmol/mol) at least 3 months after glucose-lowering pharmacotherapy stops, and it recommends continued observation. It does not test an interval (Riddle 2021, PMID 34462270). | Early sulfonylurea or insulin therapy in UKPDS still carried a microvascular relative risk reduction of 26% (95% CI 14–36; absolute 3.5%) and a mortality reduction of 10% (2–17) for up to 24 years after the glycaemic trial ended (Adler 2024, PMID 38772405). | The 2026-10-07 PubMed search did not identify a trial comparing surveillance intervals, or deintensification of complication screening, after medication-free remission. PubMed on 2026-10-07 returned 2 records (query in the log). One is a metabolic-surgery review (Dholakia 2026, PMID 41972326, title only). In an ACCORD/ACCORDION post-hoc analysis (n=2,285), people with no baseline retinopathy and mean on-treatment HbA1c below 7% had 0/386 vision-threatening events at 4 years, versus 6/721 (0.8%) with mean HbA1c of 7% or higher (P=0.54) (Tsujimoto 2021, PMID 33549254). That exposure is on-treatment HbA1c, and the interval was not randomised. | OQ-16 |
| D-13 | A narrative review found that MASLD accompanies more atherosclerosis and cardiovascular events, but did not find higher atherosclerotic cardiovascular mortality; shared metabolic risks complicate causal interpretation (Driessen 2025, PMID 38147315). | In 122 people at type 2 diabetes diagnosis, 79% had ultrasound steatosis. The investigators used FIB-4 to select intermediate- and high-risk patients for transient elastography; 25% of those referred had advanced fibrosis, a selected rather than population denominator (Barana 2024, PMID 38347338). | The cardiovascular-overlap review and the liver-triage cohort measure different outcomes. Neither tests whether coupling fibrosis triage to cardiovascular management reduces events or unnecessary referrals. | OQ-17, OQ-23 |
| D-15 | In 2023, among people with diabetes aged 15 years and older in 204 countries, 55.8% (95% uncertainty interval 49.3–62.3) were diagnosed, 91.4% (88.0–94.2) of those diagnosed were treated, and 41.6% (35.7–48.5) of those treated had optimal glycaemia. Optimal glycaemia on treatment, among everyone with diabetes, was 21.2% (17.4–25.6) (Stafford 2025, PMID 40934935). | In 55 low- and middle-income countries (680,102 adults; 37,094 with diabetes; countries weighted equally), 4.6% (95% CI 3.9–5.4) received every recommended treatment. Cholesterol-lowering coverage was 6.3% (5.5–7.2) (Flood 2021, PMID 35211689). | The 2026-10-07 PubMed search did not identify a completed cluster-randomised trial testing a diagnosis-to-control package on a population denominator with glycaemic control as the outcome. PubMed on 2026-10-07 returned 9 records (query in the log). Point-of-care capillary HbA1c raised testing uptake from 37.5% to 76.0% and diabetes detection from 1.4% to 4.2% in 852 at-risk primary-care patients; control after diagnosis was not measured (Chan 2025, PMID 40165254). A 48-village rural China cascade study is a protocol (Yin 2024, PMID 39494843). ADDITION-Cambridge invited high-risk adults to screening and found no reduction in all-cause mortality over a median 9.6 years (HR 1.06, 95% CI 0.90–1.25) (Simmons 2012, PMID 23040422). The remaining records are gestational diabetes, type 1 education, a prevalence paper, and a Cochrane screening review read by title only (Peer 2020, PMID 32470201). | OQ-19 |
| D-16 | In RISE youth (n=91; impaired glucose tolerance or type 2 diabetes diagnosed within 6 months), clamp β-cell function was lower at 12 months on treatment and at 15 months, 3 months off treatment, in both the glargine-then-metformin arm and the metformin arm (RISE Consortium 2018, PMID 29941500). After withdrawal, fasting glucose was 8.6±1.8 mmol/L and 7.8±0.7 mmol/L, and 2-hour glucose was 13.2±1.4 and 13.1±1.2 mmol/L, with a falling β-cell response (Hannon 2020, PMID 32985775). | In Ellipse (ages 10 to under 17; n=135), liraglutide up to 1.8 mg versus placebo, on metformin with or without basal insulin, had a placebo-adjusted HbA1c difference of −1.06 percentage points at 26 weeks and −1.30 at 52 weeks. There was no organ endpoint and no withdrawal phase (Tamborlane 2019, PMID 31034184). In DINAMO (ages 10–17; n=158), pooled empagliflozin changed HbA1c by −0.84% (95% CI −1.50 to −0.19) at 26 weeks; linagliptin changed it by −0.34% (−0.99 to 0.30). No severe hypoglycaemia was reported. There was no organ endpoint and no withdrawal phase (Laffel 2023, PMID 36738751). | The 2026-10-07 PubMed search did not identify a youth trial stopping a GLP-1 receptor agonist or an SGLT2 inhibitor and testing whether C-peptide preservation lasts. PubMed on 2026-10-07 returned 4 records (query in the log): a heart-failure statement, a RISE predictor paper, the RISE design paper, and a mechanistic metformin-plus-liraglutide trial in youth-onset diabetes (Dietsche 2024, PMID 37967247, title only). | OQ-6, OQ-20 |
| D-17 | In Look AHEAD, any remission versus none was associated with chronic kidney disease HR 0.67 (95% CI 0.52–0.87) in 4,132 participants and cardiovascular disease HR 0.60 (0.47–0.79) in 4,402, after adjustment for HbA1c, blood pressure, lipids, cardiovascular history, diabetes duration, and intervention arm. Remission was no diabetes medication and HbA1c below 6.5% at one time point. The abstract says the association may be affected by later changes in weight, fitness, HbA1c, and LDL cholesterol (Gregg 2024, PMID 38233592). In SOS, 284 of 586 participants were in remission at 2 years (71.9% after surgery, 16.5% with usual care). Remission was associated with mortality of 16.6 versus 26.0 per 1,000 person-years (adjusted HR 0.71, 95% CI 0.54–0.95), cardiovascular mortality sub-HR 0.54 (0.35–0.85), and 2.5 extra years of adjusted median life expectancy, over a median 26.2 years. The analysis is post hoc (Carlsson 2024, PMID 38896851). | In DiRECT, remission in the whole population was 6/89 (7%) with 0–5 kg of weight loss and 31/36 (86%) with 15 kg or more (Lean 2018, PMID 29221645). | The 2026-10-07 PubMed search did not identify a causal mediation analysis splitting the Look AHEAD or SOS outcome associations into a weight-loss path and a path specific to the remission state. PubMed on 2026-10-07 returned 0 records for remission, type 2 diabetes, mediation, and a kidney, cardiovascular, or mortality term (query in the log). | OQ-1, OQ-21 |
| D-18 | American College of Physicians guidance tells clinicians to aim for HbA1c between 7% and 8% in most adults with type 2 diabetes, to consider deintensifying pharmacotherapy below 6.5%, and to avoid an HbA1c target when life expectancy is under 10 years. The statement reviewed ADA and AACE/ACE guidelines. The abstract does not quote their numeric targets (Qaseem 2018, PMID 29507945). | Tirzepatide lowered HbA1c by 2.01 to 2.30 percentage points at 40 weeks, with glucose below 54 mg/dL in 0.2% to 1.7% of tirzepatide recipients and 0.4% of semaglutide recipients (Frías 2021, PMID 34170647). | The 2026-10-07 PubMed search did not identify a randomised trial comparing default HbA1c targets while holding an incretin or SGLT2-inhibitor strategy constant. A title-restricted PubMed query on 2026-10-07 returned 2 records (query in the log). Rodriguez 2025 replaces prandial insulin with weekly semaglutide in people already well controlled (PMID 39532398, title only). Lingvay 2023 reports glycaemic-target achievement on tirzepatide (PMID 36514843, title only). | OQ-22 |
| D-19 | A meta-analysis estimated NAFLD/MASLD in 65.33% (95% CI 62.35–68.18) of people with type 2 diabetes across 123 studies (N=2,224,144). Advanced fibrosis was 15.49% (6.99–30.99) in a separate biopsy subset of 12 studies (N=2,733), not in the full cohort (Younossi 2024, PMID 38521116). | In an individual-patient meta-analysis of 5,735 people with NAFLD, 33% of whom had type 2 diabetes, sequential FIB-4 then transient elastography had 66% sensitivity (95% CI 63–68) and 86% specificity (84–87) for advanced fibrosis, with 33% still needing biopsy under one threshold strategy. These estimates are for the mixed NAFLD sample, not a type 2 diabetes subgroup (Mózes 2022, PMID 34001645). | High disease prevalence does not establish the performance or clinical benefit of a sequential screening strategy specifically in type 2 diabetes. The prevalence meta-analysis and mixed-population test meta-analysis answer different questions. | OQ-17, OQ-23 |
| D-20 | In adults treated with basal insulin and without prandial insulin (n=175; baseline HbA1c about 9.1%), continuous glucose monitoring versus meter monitoring changed 8-month HbA1c by −0.4% (95% CI −0.8 to −0.1). Time in 70–180 mg/dL was 59% versus 43% (adjusted difference 15 percentage points, 95% CI 8–23) (Martens 2021, PMID 34077499). | The UKPDS sulfonylurea/insulin legacy microvascular relative risk reduction was 26% (95% CI 14–36), absolute 3.5%, for up to 24 years after the trial (Adler 2024, PMID 38772405). | The 2026-10-07 PubMed search did not identify a multi-year continuous-glucose-monitoring strategy trial in type 2 diabetes without prandial insulin using a microvascular, cardiovascular, or mortality endpoint. The narrower PubMed query returned 0 records on 2026-10-07. A broader query of monitoring trials that mention retinopathy, nephropathy, mortality, or cardiovascular events returned 4 (queries in the log). One of them compared intermittently scanned monitoring with fingerstick testing for about 3 months after myocardial infarction (n=141): time in range changed by +17 min/day (95% credible interval −105 to +153), time below 3.9 mmol/L fell by 80 min/day, HbA1c fell by 7 mmol/mol in both arms, and combined glycaemic emergencies plus deaths were 4 versus 7. The authors say longer-term clinical benefit is untested (Ajjan 2023, PMID 36516054). The other three titles are a drug trial that used monitoring as a measurement (PMID 41208635), a post-hoc retinal-thickness analysis (PMID 40239470), and monitoring plus low-carbohydrate coaching (PMID 35107429). | OQ-24 |
| D-21 | SURPASS-CVOT compared tirzepatide up to 15 mg with dulaglutide 1.5 mg. In the modified intention-to-treat population (6,586 versus 6,579), major cardiovascular events occurred in 12.2% versus 13.1% (HR 0.92, 95.3% CI 0.83–1.01; noninferiority p=0.003; superiority p=0.09) (Nicholls 2025, PMID 41406444). | SOUL compared oral semaglutide up to 14 mg with placebo (n=9,650; mean follow-up 47.5 months). Major cardiovascular events occurred in 12.0% versus 13.8% (HR 0.86, 95% CI 0.77–0.96). The confirmatory secondary outcomes, including the major kidney composite, did not differ (McGuire 2025, PMID 40162642). FLOW compared injectable semaglutide with placebo on a kidney primary outcome (HR 0.76, 95% CI 0.66–0.88) (Perkovic 2024, PMID 38785209). | These trials answer different comparator questions. The 2026-10-07 PubMed search did not identify a pragmatic trial randomising which agent is used first, on contemporary background therapy, with kidney or cardiovascular events as the outcome. PubMed on 2026-10-07 returned 1 record (query in the log): a retrospective comparison of SGLT2 inhibitors and semaglutide in diabetic nephropathy (Joseph 2025, PMID 40772215, title only). | OQ-3, OQ-25 |
| D-22 | A review of 27 studies recommending 53 diabetes-specific patient-reported outcomes found diabetes distress and self-management behaviours the most endorsed (Hamilton 2023, PMID 37672919). | In adults with type 2 diabetes, psychological interventions changed HbA1c by a pooled mean difference of −0.21 (95% CI −0.31 to −0.10; 49 studies, n=12,009), which the authors equate to about −0.33% or 3.5 mmol/mol. Depressive symptoms were not consistently improved (Winkley 2020, PMID 32568666). Non-specialist psychological interventions changed HbA1c by −0.13 (−0.22 to −0.04) with I²=71%, and effects on distress and depression were inconsistent across trials (Oyedeji 2022, PMID 35120528). | A routine-care randomised trial already measured physical and psychological concerns and fed them into diabetes consultations in 98 unselected adults with type 1 or type 2 diabetes (any HbA1c). HbA1c fell by 6 versus 3 mmol/mol and distress and other psychosocial measures moved favourably, but psychosocial between-group effects were not statistically significant; the abstract gives no type 2-only effect (Kelly 2025, PMID 37350136). Thus a blanket absence of randomised measurement-plus-response is false. Earlier trials tested screening plus a letter in 223 people already above a depression threshold, without a distress or HbA1c benefit (Pouwer 2011, PMID 21221528), and stepped care in 260 people selected by depression and/or distress scores, improving depression response to 80.2% versus 51.2% but not distress or HbA1c (Schmitt 2022, PMID 34875666). The remaining question is whether a prespecified distress-response pathway improves outcomes in an adequately powered type 2 diabetes-only trial. PubMed searches on 2026-10-07 returned 12 records for the narrow query and 14 for a broader routine-care screening query (both in the audit log). | OQ-8, OQ-18, OQ-26 |