Glycemic variability

In adults with T2D diagnosed within five years and no DPN at entry, reducing the amplitude of nonhypoglycemic glucose fluctuations in the peripheral sensory-neuron system, without changing daily mean exposure or low-glucose trajectories, would prevent at least 20% of five-year net physical peptidergic C-terminal loss in a baseline group containing at least 10% of reference-care incident DPN cases.

This is an incident-disease hypothesis. Established DPN, long-duration T1D and pain studies inform its biology without directly measuring this population and outcome.

Candidate baseline groups are all eligible people, those with 14-day glucose coefficient of variation ≥36%, HbA1c ≥8%, fasting triglycerides ≥150 mg/dL, BMI ≥30 kg/m², and insulin-treated people. Use these six groups individually, without additional intersections or selection by later response. Coefficient of variation is standard deviation divided by mean. These cutoffs define the scope; they are not claimed injury thresholds.

Exposure compartment and timing. Measure extracellular glucose around distal axons and terminals, their supporting Schwann cells, and the corresponding sensory-neuron somata and satellite glia. Include the tissue side of local microvessels. Systemic blood glucose is an input, not a substitute for these measurements. R1 requires a daily standard deviation ≥1.0 mmol/L among nonlow measurements on at least seven days of a 14-day period in follow-up year one. This must occur in neighborhoods serving ≥10% of mapped terminals in ≥10% of a candidate group, before an additional baseline-normalized net terminal deficit ≥5% persists for 30 days. Ordinary turnover does not constitute that sustained deficit.

Q — a defined waveform change. For each compartment and each 24-hour reference-care trajectory, preserve every concentration below 3.9 mmol/L exactly, including timing, depth and duration. For the remaining times, let m be their time-weighted mean glucose and replace each concentration g with m + 0.5(g−m). If there are no nonlow times, leave the trajectory unchanged.

This halves the standard deviation within the nonlow part of each day, preserves that day's complete glucose integral, and creates no new concentrations below 3.9. It need not halve the coefficient of variation of the entire trace. The exact low-glucose trajectory, rather than only its percentage of time below a threshold, is matched.

Apply Q to all specified neighborhoods from entry for five years. This is a controlled comparison between reference and transformed exposure trajectories, with each full trajectory available to the ideal comparison. It is not a claim that a real-time clinical controller can know a future day's mean. Prescribed external delivery supplies or removes the required glucose locally. Hold systemic glucose, insulin, other incoming fuels, oxygen delivery, temperature and osmolarity at their reference trajectories. Let local metabolism, signaling, inflammation, maintenance and repair respond naturally.

Q does not directly lower mean hyperglycemia, repair an axon or replace a missing trophic factor. Longer-lasting consequences of peaks or recovery are allowed, including changes after a low episode whose glucose depth and duration are themselves unchanged. A systemic endothelial injury route untouched by this local intervention falls outside the priced proposition.

Outcome. Follow stable-identity peptidergic terminals in fixed bilateral distal-leg and dorsal-foot territories. Physical loss means disappearance of connected terminals, minus newly established connected replacements, including subsequent loss of replacements. Peptide expression, firing and detached fragments do not identify preserved terminals. Normalize to entry terminal amount, average territories within people, then people equally. For positive reference loss L₀ and Q loss L_Q, define B=(L₀−L_Q)/L₀. Complete trajectories include losses before death; unavailable tissue observations cannot be credited as preservation.

C is the proportion of reference-care incident DPN cases belonging to a qualifying baseline group. Incident DPN requires new bilateral length-dependent neuropathy with objective small-fiber loss or abnormal nerve conduction, not pain alone. The same group must satisfy the exposure, structural-benefit and coverage requirements.

Source assessment

Best judgment: about 7%; uncertainty range (sensitivity): 0.23–51%. Smoothing glucose could reduce nerve injury even when average glucose and every low-glucose episode stay unchanged. The decisive uncertainty is whether that controlled change preserves enough human terminals; positive clinical associations do not establish this. These are subjective probabilities of the complete causal proposition, not effect sizes or a patient's neuropathy risk. Reviewed 14 September 2026. Probability method · Collection index.

Biological reasoning

If instantaneous damage is proportional to glucose and recovery has no memory, equal integrated exposure produces equal damage. A waveform effect therefore needs a nonlinear concentration response or a dependence on exposure history. Possible explanations include a concentration threshold, delayed detoxification, interrupted recovery or damage during transitions. They are alternatives; none is an additional mandatory mediator.

Even nonlinearity does not establish the sign. A saturating damaging response can make fluctuating exposure less harmful than its smoother counterpart. A beneficial response can also be lost when peaks are removed. Net physical preservation matters, rather than a transient stress marker.

A simple calculation illustrates the issue. Equal-duration levels 6 and 14 mmol/L have mean 10; Q gives 8 and 12, also mean 10. A linear damage law predicts no change. A squared-concentration law predicts only a 10.3% reduction, whereas a law counting only glucose above 12 predicts 100%. These are illustrative functions, not fitted nerve biology. Neither a variance difference nor a chosen convex curve establishes the 20% terminal-loss threshold.

Primary evidence

The strongest prospective peripheral-neuropathy counterexample is DCCT. Siegelaar analyzed 1,160 T1D participants without confirmed clinical neuropathy at baseline, with 108 incident cases after five years. Quarterly seven-point glucose profiles summarized mean exposure and within-day fluctuations over the same five-year period. After mean-glucose adjustment, the odds ratio per 1 mmol/L higher within-day standard deviation was 0.85 (95% CI 0.65–1.10); for mean amplitude of glycemic excursions, or MAGE, it was 1.00 (0.88–1.12).

Mean exposure remained associated with neuropathy. This argues against a large, easily detectable independent waveform effect in that cohort. It does not randomize variability, match every low, measure local glucose, or resolve short/nocturnal excursions using continuous monitoring. The population was T1D and the endpoint combined clinical and physiological abnormalities rather than serial terminal counts. The interval is not a bound on B under Q. Siegelaar et al., 2009.

A randomized rat comparison reaches actual skin structure, with a wide null. Kim assigned 36 rats to control, untreated streptozotocin diabetes, daily glargine, or glargine plus twice-daily glulisine. The two insulin groups had nine animals each and different within-day variability on eight-point monitoring at week 24. Epidermal fiber densities were 14.3±0.9 versus 13.9±0.6 fibers/mm, mean±SEM. The smoother-minus-variable difference is 0.40, with an approximate Welch interval −1.92 to 2.72 fibers/mm.

This excludes neither material benefit nor harm. Insulin exposure differed, final HbA1c was 12.4% versus 11.1%, and one sampling day did not establish matched six-month mean and low burden. Extra insulin or lower mean glucose could offset fluctuation injury. The body reports no significant IENFD group differences, despite a positive insulin-versus-untreated statement in the abstract. There were no serial baseline counts or peptidergic identification. Kim et al., 2022.

A prospective T2D study's positive result is pain-specific and incompletely adjusted. Chang's six-year follow-up included 662 initially DPN-free people; an age/sex-matched analysis retained 210 without DPN, 184 with nonpainful DPN and 26 with painful DPN. Highest versus lowest fasting-glucose-variability tertile gave BMI-adjusted OR 2.82 (1.04–7.64) for painful DPN, but 1.09 (0.66–1.80) for nonpainful DPN.

Variability groups also differed in mean HbA1c, fasting glucose, duration and kidney disease; the displayed outcome models adjusted only BMI after age/sex matching. Variability used visits throughout follow-up, rather than a baseline-only predictor. MNSI examination defined possible/probable DPN without electrophysiological confirmation. Across the eight displayed adjusted variability contrasts, minimum exploratory BH q is 0.28. This does not establish waveform-specific structural injury. Chang et al., 2024.

The recent large association contains numerical contradictions. Yan's retrospective study matched 128 DPN/non-DPN pairs, with concurrent inpatient glucose monitoring and motor nerve-conduction measurements. It reported MAGE 6.9 versus 4.7 mmol/L, adjusted β −0.38 (−0.52 to −0.24) for a composite conduction score, and adjusted R² 0.713. Matching HbA1c does not match monitored mean glucose or the low-glucose history.

The sensitivity analysis describes 172 cases and 140 controls, followed by 142 pairs without replacement; at most 140 pairs are possible. In the main regression, the printed unstandardized and standardized MAGE and CV coefficients imply incompatible outcome standard deviations, approximately 1.96 versus 7.91, using the reported group distributions. An unresolved unit or reporting error could explain this, but the claimed magnitude and robustness cannot be accepted as printed. The concurrent design also cannot establish direction or small-fiber loss. Yan and Wang, 2026.

Corneal evidence is exploratory. Gad measured four-day glucose profiles in 40 insulin-treated people, 20 each with T1D and T2D, plus 28 controls. Above versus below 36% variability, median corneal branch density was 25.0 versus 38.6 branches/mm², P=0.007. Within-diabetes Tables 2–3 contain 72 displayed contrasts/correlations; combined BH correction gives minimum q=0.396. Hypoglycemia and other exposure differences remain unresolved. Corneal branching is closer to physical structure than a pain score but is neither longitudinal distal-leg loss nor the controlled Q comparison. Gad et al., 2022.

Cell experiments support temporal sensitivity under severe exposures. Sun reported more Schwann-cell injury with intermittent than constant high glucose. The 5.6↔50 mmol/L, eight-hour alternation has mean 27.8, not 50. If the comparison is otherwise valid, greater injury despite lower total glucose challenges a simple monotonically increasing dose-only model; it is still not matched-mean evidence at an ordinary human nerve exposure. The accessible primary abstract does not provide recoverable donor-level contrasts or all media controls. Sun et al., 2012.

Xue's related 48-hour experiment found a 2.86-fold oxidative-probe signal with that alternation versus 5.6 mmol/L, but included no constant-high comparator. Three experiments and culture duplicates are not independent human donors. Xue et al., 2014.

Media composition can dominate the result: Yako found 15% viability at 24 hours in 50 mmol/L glucose without pyruvate, versus >90% with 1 mmol/L pyruvate, in an immortalized mouse Schwann-cell line. This constant-exposure comparison shows that substrate buffering must be addressed before transferring culture injury to a nerve waveform. Yako et al., 2021.

Conditional judgments and uncertainty

Let Eᵢ mean that at least one of the six baseline groups satisfies requirements 1 through i together. Each row prices P(Eᵢ | Eᵢ₋₁); the first prices P(E₁). A different group cannot provide each piece. Every conditional has a best estimate and a lower–upper sensitivity interval, and all endpoint costs are included in B.

See the claims and probabilities table above.

R1

Blood/interstitial fluctuations are well established. Their amplitude, filtering and timing at adult distal units remain less certain.

R2

Temporal injury is possible, but the most relevant human incident analysis is negative and the animal structural contrast is imprecise and confounded by insulin/mean glucose.

R3

Small real waveform effects need not account for one fifth of total loss. Peak thresholds could nevertheless make amplitude reduction disproportionately useful.

R4

High variability occurs in identifiable groups, but the prevalence of a material causal benefit is unknown in early T2D. A benefit confined to advanced insulin-deficient disease would fail this scope.

0.85 × 0.35 × 0.45 × 0.55 = 0.07363125, or 7.36% before rounding. The evidence does not justify distinguishing 7% from 10% as an intervention ranking.

The lower scenario [60,10,15,25]% → 0.225% treats nerve exposure as largely integrated and attributes clinical positives to mean glucose, lows, pain and disease severity. The upper [95,75,85,85]% → 51.48% assumes commonly crossed damage thresholds or slow recovery at actual local human doses. Sparse DCCT sampling and the unmatched rat treatment leave that interpretation possible.

The approximately 51-percentage-point span measures sensitivity to coherent interpretations. It is not a confidence interval estimated from participants, a fitted distribution or a claim that the rows are independent. R2 and R3 partition one physical question: their joint conditional is 15.75%. Merging them into one B≥20% row leaves the headline unchanged; measuring an oxidative marker would not settle either.

Hypothetical upward update. Replicated, mean- and low-matched local human exposure comparisons showing pre-loss engagement and B≥20%, with a measurable early-T2D case share, would move the conditionals together to [95,75,75,75]%, giving 40.1%. This addresses the physical and coverage bridges, not a larger adjusted cross-sectional coefficient.

Hypothetical downward update. Valid engagement across the candidate groups, with precise structural estimates below B=5% despite changed stress markers, would move them to [80,10,20,35]%, giving 0.56%. If only R2 were known true while all later judgments stayed fixed, the complete probability would be 21.0%; that maximum impact is not the expected value of looking for one more variability paper.

Ideal experiments that would resolve each claim

These are ideal causal comparisons, not proposed work requiring a laboratory. Shared conditions are the stated population, six fixed groups, compartments, Q, matched incoming exposures and five-year structural endpoint. Independent people are the inference units. Serial local observations establish ordering; neighborhoods and terminals do not inflate n.

For every entry, YES requires the true quantity to meet its inclusive threshold; NO requires it to fall short in every group satisfying the preceding requirements. In finite data, intervals crossing the boundary leave the claim unresolved. A failed manipulation, mismatched mean/low trajectory, selective loss of observations or unmeasured group is not evidence of NO. Use simultaneous precision across the fixed candidate family when interpreting finite observations.

1 — Does the local waveform occur before sustained injury? Simultaneously follow systemic glucose, glucose at connected compartments and stable terminal identities from entry. Compute daily nonlow SD without substituting sensor noise or a blood coefficient. YES: one group meets the ≥1 mmol/L, seven-of-14-day, ≥10% people and ≥10% terminal-serving conditions before the defined deficit. NO: adequate first-year coverage shows every group fails at least one condition. Unresolved: blood-only sampling, an unobserved distal compartment, uncertain sensor error, or injury already preceding the record. Clinical studies establish systemic variability; none supplies this whole local timing comparison.

2 — Does smoothing cause at least modest physical preservation? In R1-positive groups, compare reference trajectories with exact Q transformations. Verify daily integrals and the complete low trace in every involved compartment. Track connected loss and replacement, including harm from reduced adaptation or altered recovery. YES: one R1 group has L₀>0 and B≥5%. NO: all R1 groups have B<5% under valid engagement; zero or negative reference loss does not qualify. Unresolved: apparent benefit from lower mean glucose, extra insulin, preserved peptide expression or differential missingness. Sun supplies a possible temporal response; Kim supplies a broad but imperfect structural contrast. Neither resolves this human comparison.

3 — Is preservation material with this amplitude reduction? Use the same comparison and physical trajectories; do not introduce a stronger waveform suppression, new drug or direct growth intervention. YES: at least one R1–R2 group has B≥20%. NO: every R1–R2 group has B<20%, including genuine benefits between 5% and 20%. Unresolved: an interval straddles 20%, only one terminal-abundance time point is known, or reduced gross loss is offset by reduced replacement. This is a quantitative extension of R2 without an extra causal-mechanism requirement. No current study measures its estimand.

4 — Does a qualifying group account for enough incident disease? Ascertain baseline group membership and objective incident DPN in a representative reference-care cohort of early T2D. Link it to groups already shown to satisfy R1–R3, accounting for all cases and competing deaths. YES: at least one such group contains C≥10% of cases. NO: every such group has C<10%. Unresolved: selected inpatient case/control fractions, trial exclusions, symptom-only endpoints, a denominator from advanced diabetes, or inability to transport the same group definition. High-variability prevalence alone does not resolve causal-susceptibility prevalence.

Relation to the other mechanisms

This is a temporal modifier of glucose injury and repair. It can operate through carbonyl stress, ER signaling, oxidative injury or other routes, without making all of them necessary. It need not imply the persistent chromatin state in 40.

Hypothesis 41 concerns replacing the energetic contribution lost during defined lows. Here the low trajectory is preserved. Recovery-waveform effects can overlap without making the complete propositions identical or ordered. Rapid correction changes mean exposure over weeks; Q preserves each day's mean. Do not add these probabilities.

Available evidence cannot settle the dose-response shape in adult human peptidergic units, the net effect of this exact waveform change, or its incident-case share. Numerical support (background note outside this collection) contains the aggregate checks and illustrative response functions.