Supplementary material (v10-DEN)

Manuscript title: A Wall-Mounted Stopwatch and Propofol Dosing Patterns During Sedation Endoscopy: A Retrospective Before–After Observational Study

Target journal: Digestive Endoscopy (Original Article)

This supplement (i) provides extended background and context relocated from the compressed main text (§SA–§SD), and (ii) reports the detailed sensitivity analyses referenced in the main manuscript (§S0–§S9). Under Digestive Endoscopy rules, supporting information is unlimited; detailed material is therefore placed here rather than in the word-limited main text. All analyses use the same 2019 single-operator data extracted from the de-identified clinical record described in the main Methods. US spelling is used throughout to match the journal house style.

§SA. Extended local problem (relocated from Introduction)

From 2017 through 2018 the unit used midazolam–propofol combination sedation; in early 2019 it transitioned to propofol monotherapy. During this transition an early case produced unexpectedly deep sedation. On review, two converging decisions were identified: an additional propofol bolus had been given fewer than 30 seconds after induction because the patient was still moving — likely before the initial bolus had reached peak effect — and shortly afterward, when transient movement persisted, a further bolus was nearly administered before the operator paused; the patient settled within seconds and no further drug was required.

Both decisions shared a common difficulty: elapsed time after each bolus was hard to track at the bedside. The room contained an analogue wall clock, and the operator initially tried to anchor each bolus by memorizing the second-hand position at injection and subtracting it from the current time, but under task load a brief diversion was sufficient to lose either reference, and the residual calculation imposed an additional working-memory step at the moment a redosing decision was being made. The endoscopy console did display a stopwatch readout, but it occupied a small region of the monitor alongside the patient’s name, date of birth, and current clock time, making the elapsed-time digits hard to identify rapidly under load. This experience motivated a low-cost way to externalize elapsed time directly into the operator’s visual field during titration.

§SB. Extended rationale: cognitive aids and externalized time (relocated from Introduction)

Two principles framed the intervention. First, propofol’s therapeutic window is best treated as having two dimensions — dose (familiar) and time (under-recognized). Because premature redosing carries the principal risk of cumulative oversedation,¹¹ a cautious titration approach — starting with the lowest typical induction dose and titrating up only as observation requires — places the binding constraint on the next observation rather than the next dose, so accurate elapsed-time judgment becomes the rate-limiting step.

Second, the most reliable way to reduce predictable judgment errors under load is to externalize the relevant information into the environment, so that no additional cognitive effort is required to retrieve it. This principle is articulated in Thaler and Sunstein’s Nudge¹² and Gawande’s The Checklist Manifesto,¹³ and demonstrated empirically by the WHO Surgical Safety Checklist evaluation⁴ and a systematic review of cognitive aids in clinical emergencies.⁵ A large, visible digital stopwatch satisfies both principles for elapsed time: it is salient (the elapsed-time digits are pre-attentive in the operator’s visual field) and externalized (the operator reads the time rather than reconstructing it from memory).

§SC. Extended limitations and unmeasured confounders (relocated from Limitations)

Pre-existing temporal trend. Median WAPD was already declining in both sexes before the intervention (§S7). In the interrupted time-series adjustment the female-versus-male level shift attenuated to a value compatible with no effect. However, the female–male gap was stable across pre-intervention months (slope p = 0.51), so the level shift in the gap at July 2019 is not pre-trended in the way absolute levels are. The dose-distribution and BMI-substructure findings depend on within-period distribution rather than on attributing the level shift to the intervention alone, and are less affected by this concern.

Total-body-weight scaling. WAPD is mechanically inflated in lower-BMI patients and deflated in higher-BMI patients, and the cross-sectional female–male WAPD difference within each BMI category was small (≤ +0.085 mg/kg). The intervention-related reduction is therefore most accurately described as a redistribution of doses along the BMI gradient correlated with sex through body size, together with a within-stratum component (clearest in the obese stratum). We do not claim that equal mg/kg by total body weight is a clinically optimal target.

Underweight stratum. Of 32 underweight patients (24 female, 8 male), only 4 male patients per period were available, insufficient for within-sex inference. The female underweight stratum (n = 7/17) shows a directionally consistent reduction (median ΔWAPD −0.124) but with a wide bootstrap 95% CI [−0.21, +0.18]. Reported descriptively only.

Post-hoc framing. The initial outcome of interest was total weight-adjusted dose; differential within-sex changes observed in the data led to the sex/BMI analyses, which are exploratory and hypothesis-generating.

Absence of structured safety and depth data. No structured capture of oxygen saturation, recovery time, jaw-thrust maneuvers, sedation-depth scores, or patient-reported satisfaction was performed. No safety improvement is claimed; the inference is restricted to standardization of dose patterns.

Bolus interval not measured. The proposed mechanism — reduced reliance on subjective time perception during redosing — remains inferential; the dose-distribution findings are consistent with but do not directly measure it.

Unmeasured confounders. ASA physical status, alcohol consumption, anxiety, chronic sedative use, sleep-apnea risk, and other patient characteristics were not available in the routine data. Time-invariant operator effects are differenced out by the period × sex interaction, but operator-level gradual change cannot be ruled out and was empirically detected in the run chart.

§SD. Interpretation of the male signal (relocated from Discussion)

The comparison of weight-adjusted doses in male patients did not reach conventional significance (p = 0.095), but the direction aligns with the distribution shifts in Figure 2. The upward shift in male total dose (median 80→90 mg) does not violate established pharmacology: in both periods the weight-adjusted dose for obese men remained lower than for normal-BMI men (Table 3; obese-M 1.118→1.184 vs normal-M 1.314→1.327 mg/kg), preserving the observation that higher-body-mass patients require lower mg/kg. The increase was concentrated at higher dose levels (90 and 120 mg) rather than spread across all men, consistent with the time cue reducing cautious under-dosing rather than causing generalized oversedation. We do not claim the post-intervention pattern is optimal, only that the pre-existing sex gap narrowed.

§S0. Total-dose distribution at each 10-mg dose level (referenced from main text Figure 2 / Table 2)

Two-proportion z-test of the proportion of patients receiving each common total-dose level, comparing Before vs After within each sex. Δ is After − Before in percentage points.

Δ, After − Before change; pp, percentage points. Female 80 mg drops 18.3 pp while 70 mg rises 18.9 pp (a coordinated one-step downshift); male 80 mg drops 15.8 pp while 90 mg rises 13.6 pp (a coordinated one-step upshift while 80 mg remains modal at 28.9%).

§S1. Sex-stratified propensity-score-matched cohort

Rationale. Two patient-level imbalances were detected (main Table 1): height was modestly greater after installation (165.1 vs 163.9 cm; p = 0.020) and procedure time shorter (74 vs 78 s; p < 0.001). Monthly case-mix variation in age, height, and procedure time was detectable (Kruskal–Wallis p < 10⁻⁷ for age and procedure time), reflecting screening seasonality. Propensity-score matching (PSM) provides a robustness check not relying on linearity.

Methods. Sex-stratified 1:1 nearest-neighbor matching without replacement; propensity score by logistic regression of period on age, weight, height, procedure time; caliper 0.2 SD of the logit; balance verified by standardized mean differences (|SMD| < 0.10). The primary regression was applied to the matched cohort. Software: MatchIt (R 4.4); regression in Python statsmodels 0.14.6, HC3 SEs.

Results. Matched n = 1,052 (526/526; 289 female pairs, 237 male pairs); all covariates |SMD| < 0.05.

Supplementary Figure S1 — Density convergence in the PSM cohort. Pre-installation WAPD densities are visibly separated (female right-shifted); post-installation densities almost completely overlap.

Interpretation. PSM reproduces the primary finding (β₃ −0.105 vs −0.097) in a balanced cohort, ruling out residual case-mix imbalance as a sole explanation.

§S2. Linear regression omitting procedure time

Rationale. Procedure time differed modestly between periods and could act as a partial mediator; removing it tests for mediation.

Result. β₃ = −0.095 mg/kg (95% CI −0.145 to −0.046; p < 0.001) — essentially unchanged from primary (−0.097). The 4-second median difference is not a material mediator.

§S3. BMI substituted for weight and height

Result. β₃ = −0.095 mg/kg (95% CI −0.146 to −0.044; p < 0.001). The shift is not driven by the functional form of body-size adjustment.

§S4. Without weight (avoiding double-adjustment)

Rationale. WAPD has weight in its denominator; including weight as a covariate is partial double-adjustment.

Result. β₃ = −0.085 mg/kg (95% CI −0.140 to −0.029; p = 0.003) — 12% attenuation, direction and significance preserved.

§S5. January-2019-excluded sensitivity

Rationale. January 2019 was the first month after the midazolam-to-propofol transition (learning-curve dosing).

Result. β₃ = −0.093 mg/kg (95% CI −0.143 to −0.042; p < 0.001; n = 1,346). Within-male Mann–Whitney p shifts from 0.095 (full) to 0.032 (January-excluded). Full cohort retained as primary to avoid post-hoc selection.

§S6. Median quantile regression (τ = 0.5)

Result. β₃ = −0.125 mg/kg (95% CI −0.171 to −0.079; p < 0.001). Larger than the mean-based estimate, consistent with right-tail compression in the female distribution.

§S7. Interrupted time-series analysis

Rationale. A pre-existing downward trend in absolute WAPD requires separating pre-slope from a step-shift at the boundary.

Methods. Patient-level segmented regression with sex-specific pre-slopes, level shifts at July 2019, and post-slopes; HC3 SEs. The female:after term is the ITS analogue of the primary β₃.

Results.

The pre-period female–male gap was not trending (monthly gap slope +0.011; p = 0.51).

Supplementary Figure S2 — ITS model fit. Monthly median WAPD by sex with segmented-regression fitted lines, pre-period slopes (dashed), and the July 2019 boundary.

Interpretation. Under ITS the level shift attenuates to −0.039 (CI includes zero; p = 0.47) — the principal limitation of the headline interaction (main text). The dose-distribution and BMI-substructure findings depend on within-period patterns and are less affected. Pre-period gap stability (p = 0.51) indicates the gap shift at July is not a continuation of a pre-existing gap trend.

§S8. Sensitivity analysis summary

All seven analyses preserve the negative direction of the period × sex interaction; six of seven remain p < 0.01; ITS (§S7) attenuates to a CI including zero.

§S-BMI. BMI substructure detail (KSSO/WHO 3-category; referenced from main text Table 3 / Figure 4)

WAPD, weight-adjusted propofol dose; Δ, After − Before; CI, confidence interval; KSSO, Korean Society for the Study of Obesity; WHO, World Health Organization. WHO international underweight (<18.5) and normal (18.5–24.9) ranges with KSSO 2022 / WHO Asian-Pacific obesity threshold (≥25). Male underweight cell (n = 4 per period) too small for inference. 95% CI by stratified non-parametric bootstrap (B = 2,000).

§S-photo. Supplementary Figure S3 — installation photograph

Supplementary Figure S3 — Wall-mounted stopwatch in situ. Original photograph of the installation depicted as a line-drawing in main-text Figure 1. The endoscopist wears a surgical mask; identifying features are further attenuated by Gaussian face-region blur. The line-drawing (Figure 1) is the primary visual reference.

§S9. Software and code availability

Analyses used Python 3.12 (pandas 2.x, statsmodels 0.14.6, numpy, scipy) and R 4.4 (MatchIt). Hartigan’s dip test via the diptest R package. All analytic scripts and de-identified intermediate datasets are available from the corresponding author on reasonable request, subject to IRB approval (P01-202102-11-001).

Note on reference numbering: superscripts in §SA–§SD follow the main-text Chicago Note sequence (Surgical Safety Checklist ⁴; cognitive-aids review ⁵; propofol PK ¹¹; Nudge ¹²; Checklist Manifesto ¹³). These must be re-verified against the final main-text reference list during the editing pass.

End of supplement.