1型糖尿病高血糖管理新策略：基于因果模型与配对分析的体力活动降糖效应研究

本文通过一项创新的因果模型与配对分析研究，证实了体力活动（PA）对1型糖尿病（T1D）患者具有显著的急性降糖效果。研究纳入1546次运动事件，发现中位23分钟的运动可使血糖平均降低2.2 mmol/L，降糖效果远超静息状态，且低血糖风险极低（<2%）。该研究为将PA作为胰岛素治疗外的有效辅助手段提供了强有力的真实世界证据（RWE），并开发出基于实时葡萄糖水平与变化率的决策热图，推动个性化糖尿病管理。

　　

研究背景与意义

Achieving optimal glycaemic control in type 1 diabetes remains a significant challenge. Despite widespread use of continuous glucose monitoring (CGM) and hybrid closed-loop (HCL) systems, most users fail to reach the recommended target of ≥70% time in range (TIR; 3.9–10 mmol/L). Hyperglycaemia is usually corrected with rapid-acting insulin, which can have variable onset and duration of action, especially in the context of recent meals or activity, leading to inconsistent glucose responses. This highlights the urgent need for complementary strategies to manage hyperglycaemia.

Physical activity (PA) can serve as a non-pharmacologic strategy for hyperglycaemia management, lowering glucose through insulin-independent mechanisms and enhancing insulin sensitivity, with experimental data demonstrating significant reductions in glucose. Observational real-world studies provide some evidence of efficacy in lowering glucose levels above 10 mmol/L, but the absence of counterfactual comparators in these studies limits causal inference and undermines the application of PA as an acute glucose-lowering intervention.

Causal modelling offers a reliable way to estimate the effects of different factors in observational data by mimicking the conditions of a randomised controlled trial. Applying causal modelling to real-world PA events, while accounting for both individual and event-specific factors, can generate robust evidence regarding the impact of everyday behaviours, supporting the development of new evidence-based interventions.

This study employed a novel within-subject matched-pairs approach applied to the T1DEXI and T1DEXIP datasets, intending to provide the first causal evidence for the glucose-lowering effect of free-living PA in individuals with type 1 diabetes.

研究方法

Data comprised 1546 PA bouts of 10–30 min from 482 participants in the T1DEXI and T1DEXIP cohorts where glucose was >10 mmol/L. Each PA bout was matched [starting glucose, glucose rate of change, insulin on board (IOB) and glucose variability (CV)] to a matched non-PA period within the same individual using a weighted k-nearest neighbours algorithm (SMD <0.01).

The primary outcome was the change in sensor glucose from the start of PA to 20 min post-activity. Secondary outcomes included predictors of glucose response and rate of hypoglycaemia incidence.

Rather than relying on propensity scores, we adopted an outcome-oriented approach, prioritising matching on covariates that most strongly influenced glucose change following PA. Variables significantly associated with glucose change in univariate Pearson correlations were entered into a multivariable linear mixed-effects model with a random intercept for participant ID. Variables retained in this model were used for matching.

Each PA bout was matched to a non-PA control period of equal duration, drawn from the same participant's CGM data. We used a k-nearest neighbour (kNN) matching algorithm to pair PA bouts with non-PA bouts from the same participant. Final match quality was assessed using standardised mean differences (SMDs), with values below 0.1 considered indicative of acceptable covariate balance.

研究结果

3.1 主要结果：体力活动的降糖效果

PA [median 23 min: IQR (20, 30)] led to a mean glucose change of ?2.2 mmol/L (p < 0.001), compared with 0.3 mmol/L (p < 0.001) during matched non-PA periods (mean difference: ?1.9 mmol/L (p < 0.0001)).

In the mixed-effects model, after adjusting for relevant covariates and accounting for clustering by participant and matched bout, PA was associated with a significantly greater reduction in glucose compared with matched non-PA periods (mean difference ?1.9 mmol/L [95% CI ?2.1 to ?1.8]; p < 0.001). The mean glucose reduction following PA was ?2.2 mmol/L (95% CI ?2.4 to ?1.9); p < 0.001, compared with ?0.3 mmol/L (95% CI ?0.4 to ?0.2); p < 0.001 following matched non-PA periods.

3.2 血糖反应的预测因素

The strongest predictors of PA-induced glucose change were (in order) glucose rate of change, starting glucose, CV, duration and IOB.

The glucose-lowering effect of PA was greater in bouts with a faster rate of glucose decline at onset (per 0.1 mmol/L/min: ?0.90 mmol/L [95% CI ?1.19 to ?0.61]; p < 0.001), lower starting glucose (per 1.0 mmol/L: ?0.16 mmol/L [95% CI ?0.23 to ?0.09]; p < 0.001), higher pre-PA CV (per 1%: ?0.06 mmol/L [95% CI ?0.09 to ?0.03]; p < 0.001), longer duration (per minute: ?0.05 mmol/L [95% CI ?0.07 to ?0.03]; p < 0.001) and higher IOB at onset (per 0.1 U/kg: ?0.56 mmol/L [95% CI ?1.09 to ?0.04]; p = 0.035).

Older age (per year: ?0.01 mmol/L [95% CI ?0.02 to 0.00]; p = 0.015) and lower BMI (per 1 kg/m²: ?0.05 mmol/L [95% CI ?0.09 to ?0.01]; p = 0.012) were also associated with greater reductions.

3.3 低血糖风险

The absolute risk of hypoglycaemia during and up to 20 min after was 1.4% for PA bouts (22 out of 1546) and 0.1% for matched non-PA periods (1 out of 1546). All of the hypoglycaemic episodes occurred in the adult T1DEXI dataset.

Although the overall risk was low, PA was strongly associated with a higher likelihood of hypoglycaemia. In the fully adjusted generalised linear mixed model (binomial), PA was associated with a log-odds increase of 4.48, corresponding to an odds ratio (OR) of 88 (95% CI: 88, 89, p < 0.001). Two additional factors were significantly associated with hypoglycaemia risk: starting glucose (log-odds = ?0.57, OR = 0.56) and event duration (log-odds = 0.15, OR = 1.16).

3.4 决策热图工具

The three strongest predictors of glucose change, PA status (defined as whether 10–30 min of activity was undertaken), starting glucose level and rate of glucose change at event onset, were used to model predicted glucose change. Across all glucose levels and rates of change, PA events demonstrated greater glucose reductions.

By integrating CGM-derived glucose levels and rates of glucose change at the onset of physical activity, this tool provides a personalised estimate of the expected glucose response.

讨论与结论

This study provides strong evidence that 10- to 30-min bouts of PA significantly accelerate glucose lowering in people with type 1 diabetes compared to matched periods without PA. Using a within-subject matched-pairs design and 1546 free-living PA bouts, we found that a median of 23 min of PA, initiated when glucose was above 10.0 mmol/L, lowered glucose by an average of 2.2 mmol/L compared with just 0.3 mmol/L during matched non-PA periods.

These findings support the role of short bouts of physical activity (PA) as an adjunct to insulin for acute hyperglycaemia management, provided there is sufficient circulating insulin and no evidence of ketosis.

Causal inference methods are increasingly recognised as essential in diabetes research, particularly for generating real-world evidence where even crossover study designs may not fully address confounding or reflect everyday conditions.

Our findings provide robust real-world evidence that short bouts of PA, when used alongside insulin therapy, can rapidly lower glucose by ~2.0 mmol/L within 20 min, without increasing immediate hypoglycaemia risk.

We identified starting glucose and glucose rate of change as the strongest predictors of glucose lowering with PA. A steeper downward glucose rate of change at activity onset led to a greater reduction, while a rising trend reduced the effect.

IOB was a key factor influencing the glucose-lowering effect of PA. Higher IOB at activity onset significantly enhanced glucose reductions, consistent with physiological mechanisms whereby PA increases insulin sensitivity, accelerates subcutaneous insulin absorption and reduces renal insulin clearance.

The absence of a significant effect of activity type or intensity is both surprising and encouraging. It suggests that individuals with type 1 diabetes can engage in a wide variety of activities and still achieve clinically meaningful glucose reductions.

Personal factors also influenced the glucose-lowering effect of PA. Participants with lower BMI had greater glucose reductions, likely due to higher baseline insulin sensitivity. Those with higher HbA1c levels showed smaller reductions, suggesting that chronic hyperglycaemia and insulin resistance may blunt the effectiveness of PA.

In summary, a 20-min PA session lowered glucose by an average of ~2 mmol/L, which could be clinically translated into a simple '2-every-20' rule. The strong influence of starting glucose and glucose rate of change supports the use of heatmaps and digital tools to personalise this strategy in education and clinical care.