Peri-colonoscopy hypoglycaemia in type 2 diabetes: Temporal patterns and asymptomatic presentation

Abstract

BACKGROUND

Patients with type 2 diabetes mellitus (T2DM) undergoing colonoscopy face an elevated risk of hypoglycaemia due to prolonged fasting and bowel preparation. However, the temporal distribution of hypoglycaemic events across the peri-colonoscopy period and the prevalence of asymptomatic hypoglycaemia in this setting remain poorly characterised.

AIM

To describe the temporal distribution of hypoglycaemic events during the peri-colonoscopy period and to identify factors associated with asymptomatic hypoglycaemia.

METHODS

This retrospective cohort study included 851 hospitalised adults with T2DM who underwent colonoscopy between January 2021 and December 2025 at a tertiary hospital in China. The peri-colonoscopy period was divided into five clinical phases. Phase-specific incidence rates were calculated per 1000 patient-hours. Generalised estimating equations were used to identify factors associated with asymptomatic presentation.

RESULTS

Overall, 159 patients (18.7%) experienced 181 hypoglycaemic events, of which 54.1% were asymptomatic. The immediate post-colonoscopy phase (0-6 hours) had the highest incidence rate (6.93 per 1000 patient-hours), while the early bowel preparation phase (20:00-06:00) had the highest asymptomatic proportion (70.5%). In multivariable analysis, diabetes-related retinopathy [odds ratio (OR) = 3.19, 95%CI: 1.58-6.41, P = 0.001] and nocturnal occurrence (OR = 3.58, 95%CI: 1.83-7.00, P < 0.001) were independently associated with asymptomatic presentation, whereas level 2 hypoglycaemia was inversely associated (OR = 0.30, 95%CI: 0.12-0.76, P = 0.012). Risk stratification showed asymptomatic proportions ranging from 31.8% to 84.2% (P for trend < 0.001).

CONCLUSION

More than half of peri-colonoscopy hypoglycaemic events were asymptomatic among hospitalised patients with T2DM. Patients with retinopathy and nocturnal events may require intensified glucose monitoring regardless of symptom reporting; these findings should be interpreted in the context of inpatient colonoscopy care.

Key Words: Type 2 diabetes mellitus; Colonoscopy; Asymptomatic hypoglycaemia; Temporal distribution; Risk stratification; Peri-colonoscopy period; Glucose monitoring; Impaired awareness of hypoglycaemia

Core Tip: This study reveals that more than half (54.1%) of hypoglycaemic events during the peri-colonoscopy period in hospitalised patients with type 2 diabetes mellitus were asymptomatic, representing a substantial monitoring blind spot. Two distinct high-risk windows were identified: (1) The immediate post-colonoscopy phase (highest incidence); and (2) The early bowel preparation period (highest asymptomatic proportion). Diabetes-related retinopathy and nocturnal occurrence were independently associated with asymptomatic presentation, enabling an exploratory risk stratification from 31.8% to 84.2% asymptomatic proportions. These findings support prospective evaluation of phase-specific, risk-stratified glucose monitoring protocols in hospitalised patients.

INTRODUCTION

Colonoscopy is an essential tool for colorectal cancer screening, particularly for patients with type 2 diabetes mellitus (T2DM), who have a higher risk of colorectal cancer[1,2]. However, prolonged fasting and dietary modification during bowel preparation[3,4], together with adjustment of antihyperglycaemic therapy, create substantial challenges for glycaemic management in this population[4]. Our prior work showed that T2DM patients undergoing colonoscopy had approximately twice the risk of hypoglycaemia compared with those not undergoing colonoscopy[5]. We subsequently developed a prediction model for hypoglycaemia during the peri-colonoscopy period and identified insulin use, sodium-glucose cotransporter 2 (SGLT2) inhibitors, fasting C-peptide, and estimated glomerular filtration rate (eGFR) as key predictors[6]. However, those studies addressed the overall risk association (patient-level) and predictive modelling, respectively; neither examined the temporal distribution of hypoglycaemic events across the peri-colonoscopy course nor characterised the phenomenon of asymptomatic hypoglycaemia in this setting. The present study extends this line of research with an expanded cohort (851 patients, from January 2021 to December 2025) using event-level analysis to address these distinct and clinically important questions.

Despite increasing recognition of colonoscopy-related hypoglycaemia risk, two major knowledge gaps remain. First, previous studies have largely focused on time points relative to the procedure (before vs after colonoscopy) and have not systematically characterised finer-grained temporal patterns across the entire peri-colonoscopy course; clarifying these patterns is critical for optimising glucose monitoring workflows. Second, and more importantly, “asymptomatic hypoglycaemia” (blood glucose < 3.9 mmol/L without typical warning symptoms)[7] in this clinical context has received little attention. Asymptomatic hypoglycaemia is particularly dangerous because patients may not recognise the event, leading to delayed treatment and potentially more severe consequences[8]. Moreover, hypoglycaemia has been linked to increased cardiovascular risk, including cardiac arrhythmias and autonomic dysfunction[9], which may be particularly hazardous during procedural settings involving sedation and haemodynamic changes.

Long-standing diabetes and recurrent hypoglycaemia may result in impaired awareness of hypoglycaemia (IAH), which has been reported in 17.01% of insulin-treated patients with T2DM[10]. However, the proportion of asymptomatic events among peri-colonoscopy hypoglycaemic episodes, and the predictors of asymptomatic presentation in this specific setting, have not been systematically studied. This constitutes an important clinical blind spot: Reliance on patient-reported symptoms alone may substantially underestimate the true hypoglycaemic burden.

We hypothesised that a substantial proportion of peri-colonoscopy hypoglycaemia is asymptomatic, and that specific clinical phases and patient characteristics predict asymptomatic presentation. To test this hypothesis, we performed a comprehensive analysis: (1) Describing the temporal distribution of hypoglycaemic events across five clinically meaningful phases and comparing exposure-adjusted incidence rates; (2) Calculating phase-specific proportions of asymptomatic events; and (3) Using generalised estimating equations (GEE) to identify independent factors associated with asymptomatic presentation while accounting for recurrent events within patients.

MATERIALS AND METHODS

Study design and participants

This was a retrospective cohort study conducted at a tertiary hospital in China. We included hospitalised adults with T2DM who underwent colonoscopy between January 1, 2021 and December 31, 2025.

Inclusion criteria: (1) Age ≥ 18 years; (2) T2DM diagnosis (Tenth Revision of International Classification of Diseases code E11); and (3) Colonoscopy performed during hospitalisation.

Exclusion criteria: (1) Hospital stay < 24 hours; (2) Incomplete colonoscopy; (3) Hypoglycaemia occurring outside the predefined peri-colonoscopy period; and (4) Incomplete records for glucose values. This retrospective cohort study was approved by the Institutional Review Board (No. K2022-148-01) with informed consent waived, and was reported in accordance with the STROBE guidelines[11].

Definitions

The peri-colonoscopy period was predefined as the interval from initiation of fasting at 20:00 on the evening before colonoscopy to the time point recorded in nursing/medical notes as “resumption of a normal diet”, or the time of hospital discharge (whichever occurred first). This definition captures all hypoglycaemic events occurring within the metabolically vulnerable period, including patients with prolonged post-procedural fasting (e.g., after polypectomy).

Hypoglycaemia was defined as blood glucose < 3.9 mmol/L (70 mg/dL). Level 2 hypoglycaemia was defined as blood glucose < 3.0 mmol/L (54 mg/dL)[7]. If no hypoglycaemia-related symptoms (e.g., palpitations, sweating, tremor, confusion, or neuroglycopenic symptoms) were documented at the time of measurement, the event was classified as asymptomatic hypoglycaemia[7]; symptom information was obtained from nursing records and/or physician progress notes. Episodes were classified as symptomatic if any hypoglycaemia-related symptoms were documented, and asymptomatic only when symptoms were explicitly documented as absent (e.g., “no symptoms”). If symptom status was not documented, it was coded as undocumented and addressed in sensitivity analyses.

Phase definition

For descriptive purposes, the peri-colonoscopy period was divided into five phases: (1) Early bowel preparation: 20:00-06:00 on the night before colonoscopy; (2) Late bowel preparation: 06:00 to the start of colonoscopy; (3) Immediate post-colonoscopy: 0-6 hours after completion; (4) Early post-colonoscopy: 6-24 hours after completion; and (5) Delayed post-colonoscopy: > 24 hours after completion until resumption of a normal diet or hospital discharge.

This classification was based on three considerations: (1) Each phase corresponds to a natural transition in the clinical care workflow and nursing shift structure; (2) The metabolic milieu differs across phases owing to circadian variation in counter-regulatory hormones, laxative-induced fluid shifts, residual sedation effects, and progressive energy depletion[4]; and (3) Separating the evening-to-overnight window (20:00-06:00) from daytime preparation was essential because a substantial portion of this phase falls within the sleep period (22:00-06:00), during which patients’ ability to perceive hypoglycaemic symptoms is markedly reduced[12], which is central to the study’s focus on asymptomatic events. The 6-hour post-colonoscopy cutoff aligns with standard post-endoscopy observation periods and the typical time frame within which patients without polypectomy resume oral intake. Importantly, the five clinical phases were defined by the colonoscopy procedural timeline and are independent of the day/night classification. For example, the early bowel preparation phase (20:00-06:00) spans both nighttime hours (22:00-06:00) and the preceding evening period (20:00-22:00). The day/night classification was independently assigned based on the clock time of each hypoglycaemic event (nighttime: 22:00-06:00; daytime: 06:00-22:00) and was used as a separate predictor in the GEE analysis.

Data collection

The following information was extracted from the hospital information system: (1) Demographics (age, sex, body mass index); (2) Diabetes-related indicators [disease duration, glycated haemoglobin A1c (HbA1c), fasting C-peptide, eGFR]; (3) Microvascular complications (retinopathy, neuropathy, nephropathy); (4) Antihyperglycaemic medications (insulin, metformin, SGLT2 inhibitors); (5) Colonoscopy-related factors (procedure timing, polypectomy, post-procedural fasting duration, time to resumption of a normal diet or discharge); and (6) Hypoglycaemic event characteristics (glucose value, symptom status, timing relative to colonoscopy). In our institution, point-of-care capillary glucose testing for hospitalised patients with diabetes follows a standardised protocol: Fasting and pre-meal measurements are performed routinely [typically four times daily: (1) Before breakfast; (2) Lunch; (3) Dinner; and (4) At bedtime]. Additional measurements are performed at the discretion of the attending nurse or physician when clinical suspicion of hypoglycaemia arises. During the peri-colonoscopy period, monitoring intensity was generally consistent with this routine protocol across phases; however, the overnight bowel-preparation phase (20:00-06:00) included only a bedtime measurement, and additional testing during sleep hours was at the discretion of the attending nurse or physician. Post-colonoscopy monitoring followed standard post-procedural observation practices, with glucose checked on return from the endoscopy suite and at subsequent routine time points. Differences in monitoring intensity across phases – particularly the relative sparseness of overnight measurements compared with the immediate post-procedural period – may have influenced both incidence rates and asymptomatic proportions. All glucose measurements were obtained by point-of-care capillary testing (Rightest GM700; calibrated). Antihyperglycaemic medications were generally discontinued from the initiation of fasting and resumed according to physician orders after patients returned to a normal diet; however, individualised dose-level adjustment data – including specific insulin regimen type (basal, premixed, or intensive), dose reductions applied during fasting, and timing of resumption – were not systematically recorded. The absence of these data limits interpretation regarding the contribution of specific medication management decisions to hypoglycaemia risk, particularly given that 60.4% of hypoglycaemic patients received insulin therapy.

Statistical analysis

Continuous variables are reported as mean ± SD or median [interquartile range (IQR)], depending on distribution; categorical variables are summarised as n (%). To compare temporal clustering across phases of unequal duration, we calculated phase-specific incidence rates per 1000 patient-hours (pt-h). The pt-h were defined as the total at-risk time contributed by all participants within each phase. For phase 1 (20:00-06:00), pt-h were calculated as n × 10 hours, because fasting was protocolised to start at 20:00 for all patients. For post-procedure phases, each patient’s at-risk time was truncated at the documented time of resumption of a normal diet or hospital discharge, whichever occurred first (tend); therefore, patients who resumed eating early could contribute less than 6 hours to the 0-6 hours phase and less than 18 hours to the 6-24 hours phase.

Event-level analyses of asymptomatic presentation used GEE with a binomial distribution and logit link, specifying an exchangeable working correlation structure and clustering by patient to account for within-patient correlation from recurrent events[13]. The GEE exchangeable working correlation structure appropriately accounts for the clustering of multiple events within individual patients. Candidate predictors were prespecified based on clinical plausibility and prior literature, including retinopathy, neuropathy, nocturnal occurrence, level 2 hypoglycaemia, insulin use, diabetes duration, HbA1c, and eGFR. Univariable GEE screening (P < 0.15) was used to prioritise variables, and the final multivariable model was restricted to a small number of predictors to reduce overfitting given the limited number of asymptomatic events (n = 98).

Interaction between clinical phase and day/night classification was not included in the GEE model because the outcome of interest was asymptomatic presentation rather than hypoglycaemia occurrence, and the limited number of events within individual phase-day/night cross-classifications precluded stable estimation of interaction terms. The risk stratification was constructed post hoc based on the two strongest independent factors and should be regarded as exploratory and hypothesis-generating; a trend test was conducted using a GEE logistic model with an ordinal risk score as the predictor.

Before formal analyses, data completeness and coding consistency were checked. Key variables for hypoglycaemic events (symptom status, day/night classification, clinical phase, glucose value, retinopathy status) had no missing values. Baseline variables for the hypoglycaemia cohort (Table 1) were also complete. Therefore, the main analyses used complete-case analysis without imputation.

Table 1 Baseline characteristics of patients who experienced hypoglycaemia (n = 159), mean ± SD/n (%)/median (interquartile range).

Variable	Value
Demographics
Age (years)	58.2 ± 10.5
Male	91 (57.2)
BMI (kg/m2)	23.5 ± 3.1
Diabetes-related characteristics
Diabetes duration (years)	11 (5-18)
HbA1c [% (mmol/mol)]	8.6 ± 2.3 (70 ± 25)
Fasting C-peptide (ng/mL)	1.02 (0.51-1.69)
eGFR (mL/minute/1.73 m2)	88.8 ± 19.8
Microvascular complications
Diabetes-related retinopathy	58 (36.5)
Diabetes-related neuropathy	85 (53.5)
Diabetes-related nephropathy	19 (11.9)
Antihyperglycaemic medications
Insulin	96 (60.4)
Metformin	50 (31.4)
SGLT2 inhibitor	38 (23.9)

BMI: Body mass index; HbA1c: Glycated haemoglobin A1c; eGFR: Estimated glomerular filtration rate; SGLT2: Sodium-glucose cotransporter 2.

Asymptomatic and level 2 hypoglycaemia were defined. Nighttime was defined as 22:00-06:00 and daytime as 06:00-22:00, consistent with the typical sleep period in Chinese hospitalised patients. Post-procedure at-risk time was calculated from the colonoscopy end time (t = 0) to the endpoint time tend (hours), defined as min (Tdiet, Tdischarge): (1) 0-6 hours phase = min (tend, 6); (2) 6-24 hours phase = max [min (tend, 24) - 6, 0]; and (3) > 24 hours phase = max (tend - 24, 0). Summing across patients yielded phase-specific pt-h.

The multivariable GEE model used an exchangeable working correlation structure with robust (sandwich) standard errors. To reduce overfitting, the final model was restricted to a small number of clinically plausible predictors. Multicollinearity was assessed by variance inflation factors computed from a working-independence logistic regression; variance inflation factors values close to 1 suggested negligible collinearity.

Robustness checks included refitting the main GEE model using an independent working correlation structure and comparing effect estimates. Distributions of continuous variables were assessed visually; approximately normal variables are reported as mean ± SD, otherwise as median (IQR).

Sensitivity analyses included repeating the multivariable model using only the first hypoglycaemic event per patient (n = 159) to mitigate concerns related to recurrent events, and stratifying by biochemical severity (level 2 vs non-level 2) to evaluate whether predictors varied by severity. Two-sided P < 0.05 was considered statistically significant. Analyses were performed using R 4.3.0 and SPSS 25.0.

RESULTS

Patient characteristics

The flow of participants through the study is illustrated in Supplementary Figure 1. During the study period, 851 hospitalised patients with T2DM underwent colonoscopy; 159 (18.7%) experienced at least one hypoglycaemic event. Event-level analyses included 181 hypoglycaemic events from 159 patients; of these, 141 (88.7%) experienced a single event, 17 (10.7%) experienced two events, and one patient (0.6%) experienced six events (mean 1.14 events per patient; range 1-6). Baseline characteristics of the entire cohort, stratified by hypoglycaemia status, are provided in Supplementary Table 1. Baseline characteristics of patients who experienced hypoglycaemia are shown in Table 1: Mean age was 58.2 ± 10.5 years and 57.2% were male. Median diabetes duration was 11 years (IQR: 5-18), and mean HbA1c was 8.6 ± 2.3% (70 ± 25 mmol/mol). Diabetes-related retinopathy was present in 36.5%, and 60.4% received insulin therapy. Across the 181 events, mean glucose was 3.47 ± 0.36 mmol/L, and 11.0% (20/181) were level 2 hypoglycaemia (< 3.0 mmol/L).

Temporal distribution and incidence rates

Table 2 summarises the temporal distribution of hypoglycaemic events across the five clinical phases. The immediate post-colonoscopy phase (0-6 hours) had the highest incidence rate, followed by early bowel preparation and early post-colonoscopy phases, whereas the delayed post-colonoscopy phase had the lowest rate (Table 2). Notably, because follow-up ended at resumption of a normal diet or hospital discharge (whichever occurred first), the effective exposure duration during the 6-24 hours phase averaged 11.6 hours per patient (rather than the full 18 hours).

Table 2 Temporal distribution and incidence rates of hypoglycaemic events across five clinical phases (n = 181), n (%).

Clinical phase	Events	pt-h	Incidence per 1000 pt-h	Asymptomatic (%)	Level 2 (%)	Mean glucose (mmol/L)
Early bowel preparation (20:00-06:00)1	44 (24.3)	8510	5.17	70.5	9.1	3.46
Late bowel preparation (06:00 to start of colonoscopy)	24 (13.3)	5582	4.30	41.7	8.3	3.49
Immediate post-colonoscopy (0-6 hours)	27 (14.9)	3894	6.93	33.3	7.4	3.55
Early post-colonoscopy (6-24 hours)	49 (27.1)	9864	4.97	59.2	14.3	3.43
Delayed post-colonoscopy (> 24 hours)	37 (20.4)	14682	2.52	51.4	13.5	3.45
Total	181 (100)	42532	4.26	54.1	11.0	3.47

¹The early bowel preparation window (20:00-06:00) corresponds to the early bowel preparation period on the night before colonoscopy.

pt-h: Patient-hours (total at-risk time contributed by all participants within each phase).

Early bowel preparation showed the highest asymptomatic proportion (70.5%) despite a relatively high incidence rate, indicating a key high-risk window. In contrast, although the 0-6 hours post-colonoscopy phase had the highest incidence rate, it had the lowest asymptomatic proportion (33.3%), consistent with higher patient alertness and intensified post-procedure monitoring. Considering both incidence and asymptomatic proportion, we identified two complementary high-risk windows: (1) The immediate post-colonoscopy phase (0-6 hours) with the highest incidence; and (2) The early bowel-preparation phase (20:00-06:00) with the highest asymptomatic proportion (Figure 1).

Figure 1 Temporal distribution of hypoglycaemic events and identification of dual high-risk windows during the peri-colonoscopy period. Blue bars indicate incidence rates (left Y-axis); orange bars indicate asymptomatic proportions (right Y-axis). The two Y-axes represent different metrics and should not be interpreted as implying a direct quantitative relationship. Two high-risk windows are highlighted: (1) Early bowel preparation window (20:00-06:00) (highest asymptomatic proportion, 70.5%); and (2) Immediate post-colonoscopy (highest incidence rate, 6.93). pt-h: Patient-hours.

Overall, 54.1% (98/181) of hypoglycaemic events were asymptomatic. Level 2 hypoglycaemia accounted for 11.0% of events, with higher proportions during the early post-colonoscopy (14.3%) and delayed post-colonoscopy phases (13.5%). Importantly, within the observable inpatient period, all post-procedural hypoglycaemia events (n = 113) occurred before resumption of a normal diet. We therefore used “resumption of a normal diet or hospital discharge (whichever occurred first)” as a pragmatic endpoint to standardise the inpatient at-risk window.

Factors associated with asymptomatic hypoglycaemia

Event-level GEE models were used to identify independent factors associated with asymptomatic hypoglycaemia (vs symptomatic hypoglycaemia). In multivariable analysis, diabetes-related retinopathy [odds ratio (OR) = 3.19, 95%CI: 1.58-6.41, P = 0.001] and nocturnal occurrence (OR = 3.58, 95%CI: 1.83-7.00, P < 0.001) were independently associated with asymptomatic presentation. Level 2 hypoglycaemia was inversely associated with asymptomatic presentation (OR = 0.30, 95%CI: 0.12-0.76, P = 0.012), suggesting that more biochemically severe hypoglycaemia is more likely to present with recognisable symptoms. Clinically, these findings indicate that retinopathy may serve as a practical marker for increased odds of asymptomatic rather than symptomatic hypoglycaemia, underscoring the need for proactive glucose monitoring in this subgroup regardless of symptom reporting (Figure 2).

Figure 2 Forest plot of multivariable generalised estimating equations analysis for factors associated with asymptomatic hypoglycaemia. Orange indicates risk factors (odds ratio > 1); blue indicates protective factors (odds ratio < 1). The model used an exchangeable working correlation structure. OR: Odds ratio.

Sensitivity analysis restricted to each patient’s first hypoglycaemic event (n = 159) yielded consistent effect estimates: (1) Retinopathy OR = 3.48 (95%CI: 1.65-7.33); (2) Nocturnal occurrence OR = 3.81 (95%CI: 1.81-8.04); and (3) Level 2 hypoglycaemia OR = 0.16 (95%CI: 0.04-0.68), supporting robustness of the findings. A further sensitivity analysis restricted to daytime events (06:00-22:00, n = 109) yielded an asymptomatic proportion of 43.1% (47/109); retinopathy remained significantly associated with asymptomatic presentation (OR = 3.28, 95%CI: 1.45-7.42, P = 0.004), confirming that the observed associations are not solely attributable to nocturnal documentation practices. Symptom status was complete for all 181 hypoglycaemic events; therefore, a complete-case analysis excluding undocumented symptom status yielded identical estimates to the primary GEE model. Results of an additional robustness check using an independent working correlation structure on all 181 events were consistent with the primary analysis, with all three factors retaining statistical significance and similar effect sizes; the estimated working correlation was ρ = 0.498 (Supplementary Table 2).

Exploratory risk stratification

Based on the two strongest independent factors (retinopathy and nocturnal occurrence), we developed an exploratory, hypothesis-generating risk stratification scheme (Figure 3). The proportion of asymptomatic events increased markedly across four strata: (1) Low risk (no retinopathy + daytime), 31.8% (21/66); (2) Moderate risk B (no retinopathy + nighttime), 66.0% (35/53); (3) Moderate risk A (retinopathy + daytime), 60.5% (26/43); and (4) High risk (retinopathy + nighttime), 84.2% (16/19). A trend test indicated a highly significant gradient across risk groups (P for trend < 0.001). This simple scheme is not intended for routine clinical implementation without prospective external validation.

Figure 3 Exploratory risk stratification for asymptomatic hypoglycaemia based on retinopathy status and timing of occurrence. Asymptomatic proportions across four exploratory risk strata based on retinopathy status and timing of occurrence. This stratification is hypothesis-generating and requires prospective external validation before clinical implementation. P for trend < 0.001 (generalised estimating equations logistic model with ordinal risk score).

DISCUSSION

This study provides the first systematic characterisation of asymptomatic hypoglycaemia among hospitalised patients with T2DM undergoing colonoscopy, using “resumption of a normal diet” as a clinically meaningful anchor for the end of follow-up and analysing exposure-adjusted incidence rates. The main findings were: (1) More than half (54.1%) of hypoglycaemic events during the peri-colonoscopy period were asymptomatic; (2) Two high-risk windows were observed – an immediate post-colonoscopy window with the highest incidence rate (6.93 per 1000 pt-h) and an early bowel-preparation window (20:00-06:00) with the highest asymptomatic proportion (70.5%); and (3) Diabetes-related retinopathy and nocturnal occurrence were independently associated with asymptomatic presentation, enabling an exploratory risk stratification that requires prospective validation.

Such a high proportion of asymptomatic hypoglycaemia has important implications for clinical practice. Monitoring strategies that rely primarily on patients’ symptom reporting will inevitably miss a large share of events in this population. This finding is consistent with prior evidence on IAH in patients with long-standing diabetes[14,15], and extends the relevance of IAH to the specific context of colonoscopy preparation. The cardiovascular implications of hypoglycaemia deserve particular attention in the peri-colonoscopy context. Hypoglycaemia-induced sympathoadrenal activation may increase the risk of cardiac arrhythmias[9], which could be compounded by sedation-related haemodynamic changes during colonoscopy. These considerations further support the need for proactive glucose monitoring rather than symptom-dependent testing in this setting.

The 54.1% asymptomatic proportion observed in this study is substantially higher than the 45% reported among insulin-treated inpatients in general hospital settings[16], likely reflecting the particular features of the peri-colonoscopy setting[17], including prolonged fasting, bowel-preparation-related fluid and electrolyte shifts[18], and a narrower time window for medication adjustments[19,20]. Most previous studies have focused on glycaemic management during bowel preparation alone. By extending observation through the post-colonoscopy recovery period, our study provides a more complete assessment of hypoglycaemic burden across the entire peri-procedural course. Notably, the overall hypoglycaemia incidence of 18.7% was identical to that reported in our previous prediction-model study[6], supporting the representativeness of this cohort. It is worth noting that recent continuous glucose monitoring (CGM)-based studies in hospitalised patients with T2DM have reported substantially higher hypoglycaemia detection rates than intermittent capillary testing[21,22]; thus, the 54.1% asymptomatic proportion observed in the present study, which relied on point-of-care monitoring, likely represents a conservative estimate of the true burden.

Exposure-adjusted incidence rates provide a more rigorous metric than event counts or percentages alone. When exposure durations differ across phases and follow-up is truncated by early resumption of diet or discharge, reporting only counts or proportions can be misleading. For example, the early post-colonoscopy phase (6-24 hours) accounted for the largest proportion of events (27.1%), but its incidence rate (4.97 per 1000 pt-h) was comparable to other high-risk phases because the mean exposure in this phase was only 11.6 hours. This distinction has practical implications for allocating monitoring resources across phases.

The overnight bowel-preparation (20:00-06:00) represents an actionable prevention target: Metabolic stress from fasting and bowel cleansing may be compounded by reduced monitoring and decreased alertness during sleep, resulting in the highest asymptomatic proportion. In contrast, the immediate post-colonoscopy phase shows the highest incidence rate, potentially related to residual fasting, procedural stress, and mismatched medication adjustments. Together, these two windows suggest that a phased, targeted monitoring strategy, rather than uniform monitoring frequency throughout hospitalisation, warrants prospective evaluation.

Insulin was used by 60.4% of patients, consistent with a high-risk profile for hypoglycaemia. Our prior study identified insulin use as an independent predictor of hypoglycaemia during the peri-colonoscopy period[5,6]. In this setting, inappropriate insulin dose reduction, timing, or regimen selection may contribute to high incidence. The absence of detailed insulin regimen and dose-adjustment data is an important limitation of this study. We were unable to evaluate associations with specific insulin regimen types (e.g., basal vs premixed vs intensive therapy), nor could we assess whether individualised dose adjustments during fasting adequately mitigated hypoglycaemia risk. This limitation is particularly relevant given the high proportion of insulin-treated patients in the hypoglycaemia cohort (60.4%), and future prospective studies should systematically capture insulin management decisions across the peri-colonoscopy period to better delineate the mechanisms underlying hypoglycaemia risk in this setting. In the overall cohort, 34.9% used SGLT2 inhibitors, whereas among patients who developed hypoglycaemia the proportion was 23.9%; previous work suggests that this drug class may confer some protection against peri-colonoscopy hypoglycaemia potentially through insulin-independent glucose-lowering mechanisms[6].

Diabetes-related retinopathy was a strong correlate of asymptomatic presentation, but this association should not be interpreted as causal. Retinopathy may serve as a practical clinical marker of longer diabetes duration, autonomic dysfunction, cumulative hypoglycaemia exposure, or more advanced diabetes severity, all of which may attenuate warning symptoms of hypoglycaemia[23-25].

The association between nocturnal occurrence and asymptomatic presentation aligns with the well-recognised phenomenon that nocturnal hypoglycaemia is less readily perceived[26]. During sleep, arousal thresholds increase and caregivers’ ability to detect events decreases, making hypoglycaemia more likely to be documented as asymptomatic and potentially treated later[27].

A key methodological strength of this study is the use of a pragmatic, clinically actionable peri-colonoscopy observation window that standardises follow-up despite heterogeneous recovery trajectories. Within the observable inpatient period, all post-procedural hypoglycaemic events (n = 113) occurred before resumption of a normal diet, supporting this endpoint as an appropriate physiological-care anchor. The median time to resumption of a normal diet was 24 hours (IQR: 2-48), indicating that a substantial proportion of patients remain at risk for a prolonged post-procedure period.

From a peri-procedural management perspective, prior reviews have emphasised the need for more refined antihyperglycaemic medication adjustment and glucose monitoring around colonoscopy, with particular attention to reduced intake caused by fasting and laxatives, and to bidirectional glycaemic effects of sedation and procedural stress. These reviews also recommend bedside glucose assessment before sedation and increased monitoring frequency during high-risk periods[4,20]. Our findings further operationalise these recommendations by delineating two actionable windows – early bowel preparation (20:00-06:00) and immediate post-colonoscopy (0-6 hours) – that can be translated into phase-specific monitoring protocols.

Based on these observational findings, we propose the following stratified monitoring considerations as hypotheses for prospective validation rather than ready-to-implement clinical rules. For high-risk patients (those with retinopathy), glucose monitoring at a higher frequency (every 2-3 hours) may be considered across time periods. During early bowel preparation (20:00-06:00), even patients without retinopathy may benefit from at least 1-2 proactive glucose checks. During the immediate post-colonoscopy period (0-6 hours), glucose monitoring should be integrated into routine post-procedure assessments. In addition, nursing teams should be made aware of the high prevalence of asymptomatic hypoglycaemia to avoid relying solely on patient-reported symptoms to initiate glucose testing.

It is important to note that “asymptomatic” does not equate to “low risk”. Systematic reviews and meta-analyses indicate that IAH is not uncommon in people with diabetes and is associated with factors such as diabetes duration and prior hypoglycaemia exposure[28,29]. Therefore, among individuals with clues suggesting IAH (e.g., frequent prior hypoglycaemia or a history of nocturnal hypoglycaemia), symptom-triggered monitoring alone may be insufficient.

From a monitoring technology perspective, continuous or flash glucose monitoring has been shown to detect more hypoglycaemic events than point-of-care capillary testing, particularly nocturnal and prolonged episodes[21,30]. In the peri-colonoscopy context, CGM may help reduce nocturnal blind spots. Furthermore, older adults and patients with long disease duration (> 10 years) deserve special attention, as they often have multiple factors predisposing to impaired hypoglycaemia awareness, including autonomic dysfunction, cognitive decline, and polypharmacy; for these patients, relaxed glycaemic targets and enhanced monitoring may be prudent even without a documented diagnosis of retinopathy.

We also observed a strong association between retinopathy and asymptomatic presentation. Mechanistic and bioinformatic analyses suggest that hypoglycaemia and diabetic microvascular complications, including retinopathy, may share overlapping molecular pathways[31], and severe hypoglycaemia has been reported to predict incident retinopathy[32]. Nevertheless, in the present retrospective analysis, retinopathy should be interpreted as an associated marker rather than evidence of a direct causal pathway.

This study has limitations. First, the single-centre design may limit generalisability, although the cohort reflects common characteristics of hospitalised T2DM patients undergoing colonoscopy in China. Importantly, our findings pertain specifically to hospitalised patients and may not be directly applicable to outpatient or ambulatory colonoscopy settings, where monitoring intensity, fasting duration, and medication adjustment practices may differ. Additionally, colonoscopy preparation protocols and inpatient glucose monitoring practices vary across institutions and countries, which may affect the generalisability of these findings. Second, symptom assessment depended on nursing documentation, and nocturnal documentation may vary in consistency. Specifically, events occurring during sleep may be more likely to be classified as asymptomatic because patients were not actively queried about symptoms at the time of glucose measurement. This documentation bias could contribute to the higher asymptomatic proportion observed during the early bowel-preparation phase, and the true proportion of genuinely asymptomatic events may be lower than reported. Future studies using standardised symptom assessment protocols or CGM with alarm functions could help distinguish truly asymptomatic events from those that were simply undetected during sleep. This misclassification bias would tend to overestimate the asymptomatic proportion, and readers should interpret the reported 54.1% as an upper-bound estimate. Third, hypoglycaemia detection depended on intermittent point-of-care testing. Although institutional monitoring followed a routine protocol, actual testing frequency could differ across fasting, overnight, and post-procedure recovery periods, potentially affecting phase-specific incidence rates and asymptomatic proportions. Fourth, medication management remains incompletely characterised; detailed insulin regimens, administered doses, and dose-adjustment decisions were not systematically available, limiting interpretation of medication-related mechanisms. Fifth, post-procedure at-risk time varied because follow-up was truncated at resumption of a normal diet or hospital discharge. Although pt-h were used to account for this variability, shorter observation among patients who resumed eating or were discharged earlier may still have reduced the opportunity to detect later events. Sixth, the retrospective design precluded use of standardised instruments to assess IAH. Seventh, the absence of CGM may have led to underestimation of the true frequency of hypoglycaemia, as intermittent point-of-care testing cannot capture all glycaemic excursions, particularly brief or nocturnal episodes.

CONCLUSION

More than half of hypoglycaemic events during the peri-colonoscopy period were asymptomatic among hospitalised patients with T2DM, indicating a substantial clinical monitoring blind spot. Two high-risk windows were identified: (1) The immediate post-colonoscopy period (highest incidence); and (2) The early bowel-preparation period (highest asymptomatic proportion). Patients with retinopathy and/or nocturnal hypoglycaemic events may benefit from intensified glucose monitoring and preventive management even in the absence of typical symptoms. These findings should be interpreted primarily within inpatient colonoscopy settings and require prospective validation before broad application to outpatient practice.

ACKNOWLEDGEMENTS

We thank all the nursing staff of the Department of Endocrinology, Shenzhen Traditional Chinese Medicine Hospital, for their assistance in clinical data collection.