The decision. For an overweight or obese adult (BMI >=25), does adding a resistant-starch (RS) supplement — 10-45 g/day of a fermentable, colon-reaching starch — measurably improve blood-glucose and insulin markers? One meta-analysis speaks to this cell; nothing else on RS is held, so this page is a single-source opener (provisional), and every claim below is surrogate-level: no patient-important outcome (T2D incidence, CVD event, mortality) was measured -> Surrogate Outcomes.
Evidence state — a small, consistent surrogate signal
(Wang et al., 2019) A 2019 meta-analysis pooled 13 randomized trials, n=428, all in adults with BMI >=25; RS versus placebo/control, doses 10-45 g/day, follow-up 2-12 weeks. Effects are reported as the standardized mean difference (SMD) — a between-arm difference in pooled standard-deviation units, not a clinical unit (mmol/L, mIU/L), which limits translation to an absolute change a person would feel.
The pooled overall estimates (random-effects; a negative SMD favours RS on glucose/insulin markers):
| Marker (surrogate) | SMD (95% CI) | State |
|---|---|---|
| Fasting insulin | -0.72 (-1.13 to -0.31) | benefit |
| Fasting glucose | -0.26 (-0.5 to -0.02) | benefit (borderline, P=0.035) |
| HOMA-S% (insulin sensitivity) | +1.19 (0.59 to 1.78) | benefit (higher = more sensitive) |
| HOMA-B% (beta-cell output) | -1.2 (-1.64 to -0.77) | lower output — see interpretation |
| HbA1c | -0.43 (-0.74 to -0.13) | benefit |
| LDL-c | -0.35 (-0.61 to -0.09) | benefit |
| HOMA-IR | not significant | no meaningful effect |
| Total cholesterol | not significant | no meaningful effect |
| Triglycerides | not significant | no meaningful effect |
| HDL-c | not significant | no meaningful effect |
The abstract states the pooled result verbatim: «RS supplementation reduced fasting insulin in overall and stratified (diabetics and nondiabetics trials) analysis (SMD = –0.72; 95% CI: –1.13 to –0.31; SMD = –1.26; 95% CI: –1.66 to –0.86 and SMD = –0.64; 95% CI: –1.10 to –0.18, respectively), and reduced fasting glucose in overall and stratified analysis for diabetic trials (SMD = –0.26; 95% CI: –0.5 to –0.02 and SMD = –0.28; 95% CI: –0.54 to –0.01, respectively). RS supplementation increased HOMA-S% (SMD = 1.19; 95% CI: 0.59–1.78) and reduced HOMA-B (SMD =–1.2; 95% CI: –1.64 to –0.77), LDL-c concentration (SMD =–0.35; 95% CI: –0.61 to −0.09), and HbA1c (SMD = –0.43; 95% CI: –0.74 to –0.13) in overall analysis.» (Wang et al., 2019)
The four lipid/resistance-index nulls sit beside the significant markers: RS moved fasting insulin, fasting glucose, HbA1c, HOMA-S%, HOMA-B% and LDL-c, but not HOMA-IR, total cholesterol, triglycerides or HDL-c (Wang et al., 2019). The HOMA-IR null beside a significant HOMA-S% and falling fasting insulin is an internal wrinkle: the two indices are built from the same fasting glucose and insulin, so a consistent effect should move both. The authors do not reconcile it; read the sensitivity signal as suggestive, not settled.
HOMA-B% interpretation. A fall in modelled beta-cell output is not self-evidently good or bad. Read together with rising HOMA-S% (sensitivity) and falling fasting insulin, the coherent reading is that a more insulin-sensitive body needs less insulin output — an improvement, not beta-cell loss. But HOMA-B is a fasting-model surrogate, and its direction carries a decision only if its transmission to a patient-important outcome is evidenced; here it is not -> Surrogate Outcomes.
The diabetic-vs-nondiabetic split is a subgroup comparison, not a tested interaction
The abstract reports fasting insulin as larger in diabetics (SMD -1.26) than nondiabetics (-0.64), and fasting glucose significant only in the diabetic subgroup (Wang et al., 2019). Treat this as route-(b) effect-modification with no positive interaction evidence -> Baseline Risk and the Relative-Absolute Split:
- The subgroup CIs overlap. Diabetic -1.66 to -0.86 and nondiabetic -1.10 to -0.18 share the -1.10 to -0.86 region — two significant point estimates, no test that the effect itself differs . A larger point estimate in the higher-baseline group is exactly what route-(a) (absolute benefit scaling with baseline risk) predicts without any change in the underlying effect.
- The stratification was only possible for two markers. For HOMA-B%, HbA1c, HOMA-S%, HOMA-IR and LDL-c there was only one study per diabetic/nondiabetic group, so no stratified analysis was run (Wang et al., 2019) — the split rests on fasting glucose and fasting insulin alone.
Why the confidence is low despite a high-tier design
The design is a meta-analysis of RCTs (5 randomized-crossover + 8 parallel RCTs: «Of the thirteen trials, five of them were randomized, crossover study, the other eight were ran- domized controlled trials» (Wang et al., 2019); quality was high — 12 of the 13 studies were EPHPP-rated strong, one moderate (Wang et al., 2019)). Confidence grades total support, not design tier, and four things pull it down:
- Surrogate-only outcomes. Every endpoint is a fasting marker or lipid; none is patient-important. A marker can move the right way while the outcome does not -> Surrogate Outcomes.
- Heterogeneity was excluded, not reported. The protocol drops heterogeneous studies rather than pooling them: «If a study has a heterogeneous source, it was excluded of the analysis» (Wang et al., 2019), and it was applied — «One data were removed from analysis of the insulin and total cholesterol respec- tively because of a heterogeneous source» (Wang et al., 2019). Trimming the inconsistent arms can make a pooled effect look more consistent than the underlying literature is — the reported homogeneity is partly manufactured by the exclusion rule.
- Small, short trials. Sample sizes 12-60 per trial, 2-12 weeks (Wang et al., 2019) — too short for HbA1c to fully equilibrate (it tracks ~3-month glycaemia) and underpowered individually; the authors name «low-sample size» as «the most likely reason» for null and divergent results (Wang et al., 2019).
- Limited search. Only «PubMed and Medline» were searched (Wang et al., 2019) — no Embase, Cochrane, Scopus or trial registries — so eligible trials may be missing. Egger’s test found «no publication bias» across the nine outcomes (P from 0.153 to 0.894) (Wang et al., 2019), but that test is weak with so few trials.
Mechanism — fermentation, not the starch itself
RS is starch that resists small-intestinal digestion and reaches the colon, where bacteria ferment it. The proposed pathway is short-chain fatty acids: «SCFA, espe- cially acetate and propionate produced by colonic fer- mentation of colonic bacteria, have also been associated with the insulin sensitized effects of RS» (Wang et al., 2019), plus an animal signal on beta cells («animal models containing HAM-RS2 have shown an increase in pancreatic beta cell») and a gut-permeability/inflammation route (Wang et al., 2019). This makes RS a prebiotic fermentable fibre — the same mechanism class as inulin -> Gut Microbiome and Health, Dietary Fibre and Health. Mechanism informs direction only; the animal beta-cell result does not transport to humans on its own.
Specify the exposure — RS is not one thing
(Wang et al., 2019) The trials used different RS types (banana starch, high-amylose maize / HAM-RS2, arabinoxylan-plus-RS, RS bagels) at different doses. The authors warn: «different types of RS have opposite effects on glucose and lipid levels in healthy subjects and T2DM patients», attributing the spread to «differences in diet composition, dietary RS content, source of RS, dosage and type of RS, and the pathological status of the patients». So a recommendation must name the RS type and dose, not the label resistant starch — the pooled SMD averages over exposures that may not be equivalent -> Is the Food Category Doing Any Work.
Decision relevance — a small marginal lever, below the big rocks
For an overweight/obese person the dominant levers are weight loss, overall diet quality and activity; RS supplementation is a surrogate-level refinement that ranks below them at the margin . What it plausibly offers: a modest, well-tolerated fasting-glucose/insulin nudge — reported adverse effects were «mild and disappeared after few days of consumption» (Wang et al., 2019), reachable by swapping some refined starch for a higher-RS form or adding 10-45 g/day. What it does not yet offer: any evidenced change in a patient-important outcome, or a demonstrated effect on HOMA-IR or the lipid panel beyond LDL-c. Judge it against the realistic alternative (whole-food fermentable fibre already carries the same mechanism and its own outcome evidence -> Dietary Fibre and Health), not against placebo alone.
Gaps
- G — no hard outcome. Does RS supplementation change T2D incidence/remission or CVD events, not just fasting markers? Unmeasured here; an RS -> incidence/outcome systematic review (no such SR is held; directional-only gap).
- G — RS-type resolution. The different types have opposite effects claim needs a type-stratified synthesis (RS2 vs RS3 vs RS4) to move from label to exposure; not available in this source.
- G — durability. All trials <=12 weeks; whether the fasting-marker effect persists and whether HbA1c fully responds is untested at this horizon.