Abstract
Background
Inadequate bowel preparation (IBP) seriously compromises the quality of colonoscopy. Existing prediction models for IBP before colonoscopy lack generalizability and are insufficiently validated across diverse bowel preparation regimens. This study aimed to develop a novel model for prediction of IBP based on a systematic review and meta-analysis, and then validate its performance against those of existing prediction models in a multicenter cohort.
Methods
In the development cohort, statistically significant risk factors for IBP were systematically reviewed, and included in the novel model and weighted according to their coefficients. In the external validation cohort, the discrimination and calibration performance of the novel model was evaluated and quantified by the concordance statistic (C-statistic) and calibration slope, respectively, and then compared against existing prediction models, which were systematically reviewed and identified, in a multicenter cohort study.
Results
Twenty-five cohorts comprising 39,403 patients (11,883 with IBP) were included in the meta-analysis regarding risk factors of IBP. Age, sex, body mass index, smoking, constipation, diabetes mellitus, ASA score, history of colorectal surgery, and use of antidepressants and opioids were finally included in the novel model. By collecting 2,360 patients (445 with IBP) from 4 medical centers, its performance was externally validated [C-statistics: 0.806 and 0.744 for split-dose and full-dose 3 L polyethylene glycol (PEG) regimens], respectively; calibration slope: 1.009 and 0.985, respectively], and superior to four existing models.
Conclusion
A novel model was developed and was shown to have a favorable predictive performance of IBP in patients receiving either split-dose or full-dose 3 L PEG regimens.
Registry
ClinicalTrials.gov, TRN: NCT06438237, Registration date: 27 May 2024.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12911-026-03348-w.
Keywords: Colonoscopy, Bowel preparation, Prediction model, External validation, Meta-analysis
Background
Colonoscopy is currently the gold-standard method for evaluating the colon and screening for colorectal cancer [1]. The quality of bowel preparation significantly influences nearly all indicators of the colonoscopy procedure. Inadequate bowel preparation (IBP) is associated with reduced polyp and adenoma detection rates, prolonged cecal intubation time, increased risk of cecal intubation failure, and increased requirement of repeat colonoscopies [2]. According to the European Society of Gastrointestinal Endoscopy, the IBP rate should be maintained below 10% in patients undergoing colonoscopy [2]. However, the reported IBP rate remains as high as 18–35% [2].
Identifying patients at high risk for IBP prior to bowel preparation enables physicians to optimize bowel preparation regimens, thereby avoiding unnecessary repeat colonoscopies and reducing the risk of missed polyps and adenomas. Currently, several prediction models for IBP have been developed [3–5]. However, their generalizability is limited due to relatively small sample sizes and regional heterogeneity in study populations. Moreover, risk factors for IBP have not been comprehensively evaluated, and these models often lack external validation in patients receiving different bowel preparation regimens.
Herein, the present study aimed to develop a novel model for the prediction of IBP based on a systematic review and meta-analysis of IBP risk factors, and validate the performance of this novel model through a multicenter cohort study, with comparison to the existing prediction models.
Methods
This development and external validation study was reported in compliance with the TRIPOD statement [6] (Table S1).
Development of a novel prediction model for IBP
We performed a systematic review and meta-analysis to identify risk factors for IBP (details shown in supplementary material 1). Statistically significant risk factors identified by the meta-analysis were included in the novel prediction model. Appropriate odds ratios (ORs) and 95% confidence intervals (CIs) were selected from subgroup or sensitivity analyses to calculate the corresponding β-coefficients. The prediction model was built based on the methods suggested by Sullivan et al. [7]. The prediction was calculated by summing the scores of each predictor.
External validation
The external validation study was registered at the official website ClinicalTrials.gov (NCT06438237).
External validation cohorts
The external validation cohorts include patients from four tertiary hospitals, each utilizing a distinct bowel preparation regimen. It comprised of two parts. The first was the prospective cohort in patients undergoing colonoscopy from the General Hospital of Northern Theater Command with a regimen of split-dose 3 L polyethylene glycol (PEG), the First Affiliated Hospital of China Medical University with a regimen of 4 L PEG, and the 960th Hospital of Joint Logistic Support Force with a regimen of sodium picosulfate plus magnesium citrate (SPMC). It was approved by the Medical Ethical Committee of the General Hospital of Northern Theater Command (2024 − 104), where patients’ written informed consents were obtained during colonoscopy scheduling. The second was the retrospective cohort in inpatients undergoing colonoscopy from the Second Affiliated Hospital of Army Medical University with a full-dose 3 L PEG. In the retrospective nature of the study, the requirement for informed consent was waived by the Medical Ethical Committee of the Second Affiliated Hospital of Army Medical University (2024-188-01).
Eligible patients should be adults aged 18–94 years scheduled for colonoscopy. Exclusion criteria were as follows: (1) emergency colonoscopy; (2) cases of incomplete colonoscopy where the cause was unrelated to inadequate bowel preparation; (3) presence of major psychiatric or cognitive disorders (e.g., schizophrenia, severe dementia) preventing valid informed consent; (4) suspected intestinal obstruction, stenosis, or perforation; (5) non-compliance with bowel preparation (< 75% of prescribed dose); and (6) duplicated enrollment.
Data collection
In the prospective cohort, the data was collected at the time of colonoscopy appointment; and in the retrospective cohort, the data was collected through electronic records. The following variables were collected: bowel preparation quality, age, sex, height, weight, hospitalization status at the time of colonoscopy, comorbidities (i.e., diabetes, constipation), history of surgery (i.e., abdominal/pelvic surgery, colorectal surgery), use of medication (opioids, antidepressants), American Society of Anesthesiologists (ASA) score [8], and history of IBP which was assessed in the prospective cohort only, but not the retrospective cohort.
Predicted outcome
The predicted outcome for external validation was IBP, assessed using the Boston Bowel Preparation Scale (BBPS). IBP was defined as either a total BBPS score was < 6 or a segmental BBPS score < 2 in any colon region [9]. The BBPS score was assessed by the endoscopist during the procedure and recorded in the examination report. Endoscopists were blinded to the models’ predictive outcomes during the assessment.
Sample size calculation
Sample size was determined using the formula reported by Riley [10]. Assuming that 15.94% [11] of the recruited patients had IBP, the anticipated concordance statistic (C-statistic) was 0.8. To achieve a standard error of 0.0255 for the C-statistic and allowing for a 5% dropout rate, the sample size was set at a minimum of 552 for each cohort. Meanwhile, each cohort should include at least 100 IBP events [12].
Statistical analysis
The external validation study was performed following the steps suggested by Riley [13]. Missing data were assumed to be largely missing at random, and were imputed ten times using the mice function from the R package MICE, including all predictors and outcome. All analyses were conducted separately on each of the imputed datasets, and then the results were pooled according to Rubin’s rules [14]. Discrimination is quantified by C-statistic, which is equivalent to the area under the receiver operating characteristic curve. Other discriminatory measures, including sensitivity, specificity, positive and negative predictive values (PPV, NPV) of each model, were also calculated. The optimal cutoff value for distinguishing high-risk and low-risk patients was defined by the Youden’s index, which maximized the model’s discriminative ability for IBP prediction. Calibration was assessed using a calibration plot. Calibration should not be evaluated using the Hosmer-Lemeshow (H-L) test, because it requires arbitrary grouping of participants, which, along with sample size, can influence the calculated P value and does not indicate any miscalibration or quantify the actual magnitude. Rather, calibration should be quantified by the calibration slope (an ideal value is 1) [13]. The clinical validity and net benefit were estimated using decision curve analysis (DCA). Statistical analyses were performed with Stata software (version 18.0; Stata Corp, College Station, TX) and R 4.4.1.
Results
Development of a novel prediction model for IBP
The development cohort came from the systematic review and meta-analysis of risk factors for IBP involving 25 cohort studies. In total, 39,403 patients undergoing colonoscopy from Asia (China, Korea, Japan), Europe (Netherlands, Spain, Greece), North America (USA), and Oceania (Australia) were included in our development cohort. Among them, 11,883 patients (30.16%) had IBP. The ORs with 95% CIs for risk factors were reported or could be calculated based on the available data in all included studies. Characteristics of the development cohort are shown in Table S2.
Statistically significant risk factors identified by the meta-analysis were included in the novel prediction model (Fig. 1). The most reasonable results from subgroup or sensitivity analyses were selected, considering the feasibility of clinical practice. A significant heterogeneity was observed for gender (I²=72.38%), subgroup analysis traced its source to ethnicity, with a significant association present solely in non-Asian populations. which demonstrated a statistically significant association with IBP risk in non-Asian populations but not in Asian subgroups. Consequently, we developed two distinct versions of the prediction model: one for non-Asian patients that includes gender, and one for Asian patients that excludes it.
Fig. 1.
Pooled ORs and 95% CIs for risk factors for IBP that were statistically significant in the systematic review and meta-analysis
The risk factors ultimately included in the novel prediction model were age (with a 1-year increment) (OR = 1.008, 95% CI: 1.003–1.014, P = 0.002), male sex (only for non-Asian version, OR = 1.365, 95% CI: 1.219–1.529, P < 0.001), BMI ≥ 30 kg/m2 (OR = 1.615, 95% CI: 1.376–1.896, P < 0.001), smoking (OR = 1.260, 95% CI: 1.162–1.366, P < 0.001), constipation (OR = 2.796, 95% CI: 2.328–3.357, P < 0.001), diabetes mellitus (OR = 2.223, 95% CI: 1.811–2.728, P < 0.001), ASA score ≥ 3 (OR = 1.634, 95% CI: 1.197–2.231, P = 0.002), history of colorectal surgery (OR = 3.811, 95% CI: 2.532–5.736; P < 0.001), use of antidepressants (OR = 4.670, 95% CI: 3.106–7.023, P < 0.001), and use of opioids (OR = 1.602, 95% CI: 1.193–2.151, P = 0.002) (Fig. 2). Scores for each risk factor were further determined (Table S3). Subsequently, they were used to develop a score system for predicting IBP (Table 1).
Fig. 2.
Pooled ORs and 95% CIs for subgroup and sensitivity analyses for IBP risk factors eligible for model development
Table 1.
The novel IBP prediction model
| Risk factors for IBP | Categories | Scores |
|---|---|---|
| Age a | ||
| 18–44 | -1 | |
| 45–69 | 0 | |
| 70–94 | 1 | |
| Sex b | ||
| Female | 0 | |
| Male | 1.5 | |
| BMI | ||
| BMI < 30 kg/m2 | 0 | |
| BMI ≥ 30 kg/m2 | 2.5 | |
| Smoking | ||
| No | 0 | |
| Yes | 1 | |
| Constipation | ||
| No | 0 | |
| Yes | 5 | |
| Diabetes mellitus | ||
| No | 0 | |
| Yes | 4 | |
| ASA score ≥ 3 | ||
| No | 0 | |
| Yes | 2.5 | |
| History of colorectal surgery | ||
| No | 0 | |
| Yes | 6.5 | |
| Use of antidepressants | ||
| No | 0 | |
| Yes | 7.5 | |
| Use of opioids | ||
| No | 0 | |
| Yes | 2.5 |
The total score was calculated by summing the scores of each predictor. For individuals, the higher the cumulative score, the higher the risk of IBP
BMI, body mass index; ASA, American Society of Anesthesiologists
a Patients in development cohorts aged 18–94 years. b Only applicable to non-Asian individuals
Characteristics of the external validation cohort
The external validation cohort comprised a total of 2,360 patients, including 552 in the split-dose 3 L PEG cohort, 552 in the full-dose 3 L PEG cohort, 660 in the 4 L PEG cohort, and 596 in the SPMC cohort (Figure S1). Characteristics of each cohort are summarized in Table 2.
Table 2.
Characteristics of external validation cohorts
| Characteristics | All participants n = 2360 |
Split-dose 3 L PEG cohort n = 552 |
Full-dose 3 L PEG cohort n = 552 |
4 L PEG cohort n = 660 |
SPMC cohort n = 596 |
|---|---|---|---|---|---|
| Age [median (IQR)] | 54 (42–63) | 54 (42–62) | 55 (46–67) | 55 (44–63) | 51 (37–62) |
| Sex | |||||
| Male | 1262 (53.5) | 293 (53.1) | 256 (53.7) | 341 (51.7) | 372 (62.4) |
| Female | 1098 (46.5) | 259 (46.9) | 296 (46.3) | 319 (48.3) | 224 (37.6) |
| Height [median (IQR)] | 168 (160–175) | 169 (162–176)a | 160 (155–167) | 169 (161–179) | 170 (162–175) |
| Weight [median (IQR)] | 66 (57–75) | 67 (58–78)b | 59 (51–66) | 68 (57–77) | 59 (60–78) |
| BMI [median (IQR)] | 23.45 (21.11–25.53) | 23.39 (20.83–26.37) | 22.68 (20.46–25.30) | 23.33 (21.24–24.91) | 24.22 (21.95–26.45) |
| Inpatients | |||||
| Yes | 1240 (52.5) | 153 (27.7) | 552 (100.0) | 128 (19.4) | 407 (68.3) |
| No | 1120 (47.5) | 399 (72.3) | 0 | 532 (80.6) | 189 (31.7) |
| Smoking | |||||
| Yes | 679 (28.8) | 114 (20.7) | 152 (27.5) | 250 (37.9) | 163 (27.3) |
| No | 1681 (71.2) | 438 (79.3) | 400 (72.5) | 410 (62.1) | 433 (72.7) |
| Constipation | |||||
| Yes | 313 (13.3) | 91 (16.5) | 49 (8.9) | 89 (13.5) | 84 (14.1) |
| No | 2047 (86.7) | 461 (83.5) | 503 (91.1) | 571 (86.5) | 512 (85.9) |
| Diabetes mellitus | |||||
| Yes | 304 (12.9) | 70 (12.7) | 84 (15.3) | 79 (12.0) | 71 (11.9) |
| No | 2056 (87.1) | 482 (87.3) | 468 (84.8) | 581 (88.0) | 525 (88.1) |
| Co-morbidity | |||||
| Yes | 393 (16.7) | 91 (16.5) | 128 (23.4) | 85 (12.9) | 89 (14.9) |
| No | 1967 (83.3) | 461 (83.5) | 424 (76.6) | 575 (87.1) | 507 (85.1) |
| ASA score | |||||
| <3 | 2210 (93.6) | 536 (97.1) | 488 (88.4) | 617 (93.5) | 569 (95.5) |
| ≥3 | 150 (6.4) | 16 (2.9) | 64 (11.6) | 43 (6.5) | 27 (4.5) |
| Use of antidepressants | |||||
| Yes | 51 (2.2) | 15 (2.7) | 3 (0.5) | 21 (3.2) | 12 (2.0) |
| No | 2309 (97.8) | 537 (97.3) | 549 (99.5) | 639 (96.8) | 584 (98.0) |
| Use of opioids | |||||
| Yes | 79 (3.3) | 16 (2.9) | 25 (4.5) | 17 (2.6) | 21 (3.5) |
| No | 2281 (96.7) | 536 (97.1) | 527 (95.5) | 643 (97.4) | 575 (96.5) |
| History of inadequate bowel preparation c | |||||
| Yes | 87 (4.8) | 23 (4.2) | - | 43 (6.5) | 21 (3.5) |
| No | 1721 (95.2) | 529 (95.8) | - | 617 (93.5) | 575 (96.5) |
| History of abdominal / pelvic surgery | |||||
| Yes | 310 (13.1) | 72 (13.0) | 83 (15.0) | 86 (10.2) | 69 (11.6) |
| No | 2050 (86.9) | 480 (87.0) | 469 (85.0) | 593 (89.8) | 527 (88.4) |
| History of colorectal surgery | |||||
| Yes | 235 (10.0) | 46 (8.3) | 77 (14.0) | 67 (10.2) | 45 (7.6) |
| No | 2125 (90.0) | 506 (91.7) | 475 (86.0) | 593 (89.8) | 551 (92.4) |
| Colonoscopy time | |||||
| Morning | 1661 (70.4) | 419 (75.9) | 338 (61.2) | 377 (57.1) | 527 (88.4) |
| Afternoon | 699 (29.6) | 133 (24.1) | 214 (38.8) | 283 (42.9) | 69 (11.6) |
| Sedated colonoscopy | |||||
| Yes | 1368 (58.0) | 298 (54.0) | 396 (71.7) | 263 (39.8) | 411 (69.0) |
| No | 992 (42.0) | 254 (46.0) | 156 (28.3) | 397 (60.2) | 185 (31.0) |
| Inadequate bowel preparation d | |||||
| Total | 445 (19.1) | 100 (18.4) | 142 (26.6) | 101 (15.3) | 102 (17.1) |
| Right colonic region | 366 (15.7) | 99 (18.2) | 107 (20.0) | 100 (15.2) | 60 (10.1) |
| Transverse colonic region | 115 (4.9) | 36 (6.6) | 37 (6.7) | 10 (1.5) | 41 (6.9) |
| Left colonic region | 137 (5.9) | 24 (4.4) | 45 (8.2) | 14 (2.1) | 61 (10.2) |
| Polyps and/or adenomas detection | |||||
| Yes | 1236 (52.3) | 303 (54.9) | 200 (36.2) | 333 (50.5) | 400 (67.1) |
| No | 1124 (47.6) | 249 (45.1) | 352 (63.8) | 327 (49.5) | 196 (32.9) |
PEG, polyethylene glycol; SPMC, sodium picosulfate plus magnesium citrate; BMI, body mass index; ASA, American Society of Anesthesiologists
a One missing value. b One missing value. c History of inadequate bowel preparation was not available in the full-dose 3 L PEG cohort since it was not recorded in the medical record.d There were 10 and 18 missing values in split-dose and full-dose 3 L PEG cohort, respectively
External validation of the novel model
The C-statistic value of this novel model was 0.768 (95% CI: 0.743–0.793), and its sensitivity, specificity, PPV, and NPV were 75.00%, 67.09%, 34.33%, and 92.13%, respectively, in all participants. A cut-off score of 3.25 was selected, yielding a sensitivity of 0.694 and a specificity of 0.717. It showed good discrimination performance except in full-dose 3 L PEG cohort, with a C-statistic value of > 0.75 (Table 3; Fig. 3). Its calibration slope was 0.917 in all participants, indicating good calibration performance. In the split-dose and full-dose 3 L PEG cohorts, the calibration slopes were 1.009 and 0.958, respectively, indicating good calibration performance; however, in the 4 L PEG and SPMC cohorts, the calibration slopes were 0.780 and 0.552, respectively, indicating unsatisfactory calibration performance (Table 3; Fig. 4).
Table 3.
Discrimination and calibration performance of the novel and four existing prediction models
| Model | Performance measure | All participants n = 2360 |
Split-dose 3 L PEG cohort n = 552 |
Full-dose 3 L PEG cohort n = 552 |
4 L PEG cohort n = 660 |
SPMC cohort n = 596 |
|---|---|---|---|---|---|---|
| Wang | Discrimination | |||||
| C statistic (95% CI) | 0.768 (0.743–0.793) | 0.806 (0.759–0.853) | 0.744 (0.689–0.799) | 0.765 (0.717–0.814) | 0.762 (0.715–0.809) | |
| Sensitivity | 75.00% | 0.772 | 0.880 | 0.822 | 0.833 | |
| Specificity | 67.09% | 0.752 | 0.493 | 0.640 | 0.666 | |
| PPV | 34.33% | 41.05% | 37.09% | 29.23% | 34.00% | |
| NPV | 92.13% | 93.65% | 91.72% | 93.21% | 95.09% | |
| Calibration | ||||||
| Calibration slope | 0.917 | 1.009 | 0.958 | 0.780 | 0.522 | |
| Calibration intercept | 0.040 | -0.063 | 0.081 | 0.072 | 0.149 | |
| Dik | Discrimination | |||||
| C statistic (95% CI) | 0.736 (0.705–0.767) | 0.695 (0.640–0.751) | - | 0.783 (0.735–0.830) | 0.729 (0.673–0.785) | |
| Sensitivity | 63.49% | 0.812 | - | 0.713 | 0.539 | |
| Specificity | 74.07% | 0.483 | - | 0.751 | 0.864 | |
| PPV | 33.10% | 26.03% | - | 34.12% | 45.08% | |
| NPV | 90.94% | 91.98% | - | 93.54% | 90.08% | |
| Calibration | ||||||
| Calibration slope | 0.542 | 0.248 | - | 0.693 | 0.717 | |
| Calibration intercept | 0.104 | 0.182 | - | 0.071 | 0.057 | |
| Gimeno | Discrimination | |||||
| C statistic (95% CI) | 0.634 (0.606–0.662) | 0.673 (0.612–0.730) | 0.584 (0.525–0.642) | 0.663 (0.608–0.718) | 0.616 (0.562–0.670) | |
| Sensitivity | 49.27% | 0.594 | 0.352 | 0.554 | 0.549 | |
| Specificity | 76.28% | 0.754 | 0.812 | 0.784 | 0.694 | |
| PPV | 32.27% | 35.09% | 40.86% | 31.64% | 27.05% | |
| NPV | 86.76% | 89.24% | 77.20% | 90.68% | 88.17% | |
| Calibration | ||||||
| Calibration slope | 0.673 | 0.479 | 2.197 | 1.592 | 1.167 | |
| Calibration intercept | 0.115 | 0.211 | -0.428 | -0.084 | 0.048 | |
| Zhang | Discrimination | |||||
| C statistic (95% CI) | 0.714 (0.698–0.755) | 0.781 (0.728–0.835) | 0.716 (0.661–0.771) | 0.666 (0.605–0.727) | 0.688 (0.630–0.745) | |
| Sensitivity | 59.47% | 0.743 | 0.889 | 0.574 | 0.578 | |
| Specificity | 75.45% | 0.752 | 0.469 | 0.739 | 0.759 | |
| PPV | 35.71% | 40.11% | 38.25% | 28.43% | 33.15% | |
| NPV | 89.03% | 92.88% | 91.95% | 90.57% | 89.71% | |
| Calibration | ||||||
| Calibration slope | 0.971 | 1.030 | 0.914 | 0.851 | 0.572 | |
| Calibration intercept | -0.002 | -0.023 | 0.100 | 0.063 | 0.130 | |
| Chen | Discrimination | |||||
| C statistic (95% CI) | 0.634 (0.606–0.662) | 0.624 (0.568–0.681) | 0.573 (0.513–0.634) | 0.689 (0.640–0.739) | 0.652 (0.594–0.709) | |
| Sensitivity | 26.46% | 0.752 | 0.222 | 0.851 | 0.333 | |
| Specificity | 92.76% | 0.415 | 0.914 | 0.422 | 0.925 | |
| PPV | 45.61% | 22.35% | 48.98% | 21.03% | 47.89% | |
| NPV | 84.61% | 88.21% | 76.07% | 94.02% | 87.05% | |
| Calibration | ||||||
| Calibration slope | 1.205 | 1.960 | 1.816 | 0.708 | 1.077 | |
| Calibration intercept | -0.027 | -0.176 | -0.204 | 0.055 | 0.007 |
PEG, polyethylene glycol; SPMC, sodium picosulfate plus magnesium citrate; PPV, positive predictive value; NPV, negative predictive value
Fig. 3.
Receiver operating characteristic curve for the novel and existing prediction models in different cohorts. (A) All participants; (B) Split-dose 3 L PEG cohort; (C) Full-dose 3 L PEG cohort; (D) 4 L PEG cohort; (E) SPMC cohort. Abbreviations: PEG, polyethylene glycol; SPMC, sodium picosulfate plus magnesium citrate
Fig. 4.
Calibration plots for for the novel and existing prediction models in different cohorts. (A) All participants; (B) Split-dose 3 L PEG cohort; (C) Full-dose 3 L PEG cohort; (D) 4 L PEG cohort; (E) SPMC cohort. Abbreviations: PEG, polyethylene glycol; SPMC, sodium picosulfate plus magnesium citrate
Comparison of predictive performance between the novel and existing models
A systematic review of existing prediction models was conducted to comprehensively summarize the existing models (details shown in supplementary material 1). Among the 17 existing models, only four were available for further analysis. Reasons for excluding models for external validation were shown in Table S4.
In all participants, the novel model had the best discrimination performance, with a C-statistic value of 0.768 (95% CI: 0.743–0.793), while the performance of the existing models was inferior with C-statistic values of 0.634–0.736. While both our (calibration slope = 0.917) and Zhang’s [5] (calibration slope = 0.971) models demonstrated good calibration performance in all participants, other models showed poor calibration (calibration slope < 0.8 or > 1.2). In the two 3 L PEG cohorts, the novel model achieved superior discrimination and calibration performance. Notably, the Dik’s model [3] could not be validated in the full-dose 3 L PEG cohort, because it incorporated “history of IBP” as a predictor, which was not recorded in our medical records. In the 4 L PEG cohort, the Dik’s model [3] achieved superior discrimination performance, with a C-statistic value of 0.783 (95% CI: 0.735–0.830). However, calibration was suboptimal in all five models, with calibration slopes outside the acceptable range (0.9–1.1). In the SPMC cohort, the novel model achieved superior discrimination performance, with a C-statistic value of 0.762 (95% CI: 0.715–0.809), but it showed poor calibration performance (calibration slope = 0.522). Chen’s model [22] achieved superior calibration performance (calibration slope = 1.077), but poor discrimination performance (C-statistic = 0.652, 95% CI: 0.594–0.709) (Table 3; Figs. 3 and 4). DCA for the novel and existing prediction models in different cohorts were shown in Figure S2.
Discussion
In this study, we developed a novel prediction model for IBP based on a systematic review and meta-analysis of IBP risk factors and externally validated its performance based on a multicenter study. Additionally, the novel model showed superior discriminative ability as compared to existing models, along with good calibration performance.
Among the existing prediction models for IBP, the assessment of risk factors often lacked comprehensiveness, because the predictor selection was primarily based on univariable analyses [3–5, 15–22], unsystematic literature review [23], or prior clinical knowledge [24, 25]. To address these limitations, we conducted a systematic review and meta-analysis to rigorously evaluate patient-related IBP risk factors. We specifically included only cohort studies with high patient compliance to bowel preparation regimens, thereby minimizing confounding by preparation-related factors. Using quantitative synthesis, we derived pooled ORs to obtain robust effect estimates. This approach allowed for a more comprehensive analysis of IBP risk factors and enabled the development of a more reliable risk assessment model than those derived from individual studies.
This model is recommended for adults aged 18–94 years, consistent with the age range of our development cohort. Notably, dietary factors and preparation-to-colonoscopy interval were excluded from the model developmentas these variables could only be assessed after bowel preparation initiation and had inconsistent definitions across studies. Significant heterogeneity was observed among the 14 studies evaluating gender as a risk factor (I²=72.38%). Subgroup analysis revealed a statistically significant association between male sex and IBP risk in the non-Asian population (OR = 1.365, 95% CI: 1.219–1.529, P < 0.001), but not in the Asian subgroup. Consequently, the final prediction model for non-Asian patients includes 10 risk factors, while that for Asian patients excludes gender and retains the remaining factors. Both BMI ≥ 24 kg/m² and BMI ≥ 30 kg/m² were identified as IBP risk factors, but BMI ≥ 30 kg/m² was ultimately selected for the final model due to substantial heterogeneity among studies examining BMI ≥ 24 kg/m² (I²=87.12%) that persisted despite sensitivity and subgroup analyses, largely attributable to the limited number of included studies. This heterogeneity primarily stemmed from significant population differences between the studies that Xu et al. [26] exclusively enrolled elderly patients with a mean age of 66.49 years, while Zhang et al. [5] included predominantly younger adults with 70.4% of participants under 60 years and a mean age of 50.80 years. Furthermore, the generalizability of findings for BMI ≥ 24 kg/m² was limited as both studies involved exclusively Chinese populations, whereas BMI ≥ 30 kg/m² demonstrated more consistent evidence across diverse populations.
The ten predictors incorporated in our final model can be systematically categorized based on their underlying physiological mechanisms. Diabetes-induced autonomic neuropathy [27], the anticholinergic effects of antidepressants [28], and the µ-receptor-mediated suppression of neurotransmission by opioids [29] may directly impair intestinal motility. Advancing age is associated with a natural decline in intestinal neuromuscular function and reduced daily activity. An elevated BMI may correlate with a sedentary lifestyle. And smoking activates the sympathetic nervous system and inhibits parasympathetic activity [30]. These factors may indirectly impair intestinal motility through systemic effects on the patient’s entire body. Additionally, a history of colorectal surgery can lead to physical alterations and adhesions that may result in poor intestinal motility, ultimately diminishing bowel preparation quality. A high ASA score serves as a comprehensive indicator of overall health deterioration and may reflects a physiological milieu predisposed to inadequate bowel preparation. Furthermore, male sex has been identified as a predictor in non-Asian populations. This may because that women are more attentive to bowel preparation details during colonoscopy appointments and more frequently confirm dietary restrictions and laxative timing with healthcare staff [5].
This novel model offers essential guidance for clinical practice. It is recommended that patients with a model score higher than 3.25 receive an enhanced bowel preparation regimen, which may include increased laxative dosages, administration of adjunctive medications, and enhanced patient education. Such a risk-adapted strategy could help to improve the quality of bowel preparation, thereby improving polyp and adenoma detection rates and reducing the necessity for repeat colonoscopies.
To our knowledge, this study represents the most comprehensive external validation of existing IBP prediction models to date. We performed external validation of our newly developed model alongside four existing prediction models across four independent cohorts, each employing a different commonly used bowel preparation regimen. Among patients receiving either split-dose or full-dose 3 L PEG, the novel model demonstrated optimal performance in both calibration and discrimination. For patients using a split-dose 4 L PEG regimen, the Dik ‘s model [3] exhibited superior discriminative ability, but all models showed suboptimal calibration. In the SPMC cohort, the novel model also achieved better discrimination, while the Chen’s model [17] showed better calibration performance.
This study has several strengths. First, the study developed a novel prediction model using data from 25 cohort studies, including 39,403 patients and 30.16% of them had IBP (11,883 patients). Risk factors significantly associated with IBP were selected as predictors for the model, which addressed common limitations of individual studies, such as incomplete predictor selection and limited generalizability. Second, we systematically evaluated existing prediction models using the PROBAST criteria and compared their performance with our novel model through external validation on four independent datasets, enabling unbiased assessment of predictive ability. Third, we performed external validation across four cohorts representing different bowel preparation regimens, which were commonly used in clinical practice (split-dose 3 L PEG, full-dose 3 L PEG, 4 L PEG, and SPMC). Certainly, it also has limitations. First, the full-dose 3 L PEG cohort data were retrospectively collected from inpatient medical records, which may limit generalizability to outpatient populations. Second, three existing models were not compared in the external validation study, because they included the predictors that were not routinely collected in clinical practice, reflecting their limited real-world applicability. Third, the final predictive model has two versions: one for Asian and one for non-Asian populations. Since our validation was conducted exclusively in an Asian population, the non-Asian version was not empirically tested. This represents a limitation for the generalizability of our findings, and the performance of this version requires validation in future external studies involving non-Asian cohorts. Fourth, our predictive model was not validated in patients using novel low-volume regimens. This is because these regimens are not currently recommended by Chinese national guidelines and were not routinely used in our clinical practice during the study period. Therefore, the model’s performance in settings where these regimens are standard remains uncertain and should be explored in future studies.
Conclusion
A novel prediction model for IBP was developed based on a systematic review and meta-analysis. Predictors were age, male sex, BMI ≥ 30, smoking, constipation, diabetes mellitus, ASA score ≥ 3, history of colorectal surgery, and use of antidepressants and opioids. The novel model demonstrated superior predictive performance compared with existing prediction models in patients receiving either split-dose or full-dose 3 L PEG regimens.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
Not applicable.
Abbreviations
- IBP
Inadequate bowel preparation
- BBPS
Boston Bowel Preparation Scale
- OR
Odds ratio
- CI
Confidence interval
- BMI
Body mass index
- PEG
Polyethylene glycol
- SPMC
Sodium picosulfate plus magnesium citrate
- ASA
American Society of Anesthesiologists
- PPV
Positive predictive value
- NPV
Negative predictive value
- H-L
Hosmer-Lemeshow
- ROB
Risk of bias
Author contributions
(1) Conception and design: Weiyi Wang, Xingshun Qi, Caiping Song; (2) data collection: Weiyi Wang, Libo Tong, Shiyang Li, Wei He, Jinyuan Huang, Xiaofeng Liu, Cui Wang; (3) analysis and interpretation of data: Weiyi Wang, Junqi Xia, Xingshun Qi; (4) drafting of the article: Weiyi Wang, Xingshun Qi; (5) critical revision of the article for important intellectual content: Weiyi Wang, Libo Tong, Shiyang Li, Xingshun Qi, Caiping Song; (6) statistical analysis: Weiyi Wang, Junqi Xia; (7) final approval of the article: Weiyi Wang, Libo Tong, Shiyang Li, Wei He, Jinyuan Huang, Xiaofeng Liu, Cui Wang, Junqi Xia, Xingshun Qi, Caiping Song.
Funding
This study was supported by the Chongqing Science and Health Joint Research Project [grant number 2025MSXM124].
Data availability
The datasets analyzed during the current study are not publicly available due to ethical considerations, but are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
This study was conducted in accordance with the Declaration of Helsinki and was approved by Medical Ethical Committee of the General Hospital of Northern Theater Command (2024 − 104) and Medical Ethical Committee of the Second Affiliated Hospital of Army Medical University (2024-188-01). Patients’ written informed consents were obtained during colonoscopy scheduling in the prospective cohorts. The requirement for informed consent was waived by the Medical Ethical Committee of the Second Affiliated Hospital of Army Medical University in the retrospective cohort.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Weiyi Wang, Libo Tong and Shiyang Li contributed equally to this work.
Contributor Information
Xingshun Qi, Email: xingshunqi@126.com.
Caiping Song, Email: scp1974@163.com.
References
- 1.Hassan C, Bretthauer M, Kaminski M, et al. Bowel Preparation for colonoscopy: European society of Gastrointestinal endoscopy (ESGE) guideline. Endoscopy. 2013;45:142–55. [DOI] [PubMed] [Google Scholar]
- 2.Hassan C, East J, Radaelli F, et al. Bowel Preparation for colonoscopy: European society of Gastrointestinal endoscopy (ESGE) Guideline – Update 2019. Endoscopy. 2019;51:775–94. [DOI] [PubMed] [Google Scholar]
- 3.Dik VK, Moons LMG, Hüyük M, et al. Predicting inadequate bowel Preparation for colonoscopy in participants receiving split-dose bowel Preparation: development and validation of a prediction score. Gastrointest Endosc. 2015;81:665–72. [DOI] [PubMed] [Google Scholar]
- 4.Gimeno-García A, Baute J, Hernandez G, et al. Risk factors for inadequate bowel preparation: a validated predictive score. Endoscopy. 2017;49:536–43. [DOI] [PubMed] [Google Scholar]
- 5.Zhang N, Xu M, Chen X. Establishment of a risk prediction model for bowel Preparation failure prior to colonoscopy. BMC Cancer. 2024;24:341. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Collins GS, Reitsma JB, Altman DG, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594–7594. [DOI] [PubMed] [Google Scholar]
- 7.Sullivan LM, Massaro JM, D’Agostino RB. Sr. Presentation of multivariate data for clinical use: the Framingham study risk score functions. Stat Med. 2004;23:1631–60. [DOI] [PubMed] [Google Scholar]
- 8.Anesthesiologists ASo. New classification of physical status. Anesthesiology. 1963;24:111. [Google Scholar]
- 9.Lai EJ, Calderwood AH, Doros G, et al. The Boston bowel Preparation scale: a valid and reliable instrument for colonoscopy-oriented research. Gastrointest Endosc. 2009;69:620–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Riley RD, Snell KIE, Archer L, et al. Evaluation of clinical prediction models (part 3): calculating the sample size required for an external validation study. BMJ. 2024;384:e074821. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Feng L, Guan J, Dong R, et al. Risk factors for inadequate bowel Preparation before colonoscopy: A meta-analysis. J Evidence-Based Med. 2024;17:341–50. [DOI] [PubMed] [Google Scholar]
- 12.Wolff RF, Moons KGM, Riley RD, et al. PROBAST: A tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170:51–8. [DOI] [PubMed] [Google Scholar]
- 13.Riley RD, Archer L, Snell KIE, et al. Evaluation of clinical prediction models (part 2): how to undertake an external validation study. BMJ. 2024;384:e074820. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Marshall A, Altman DG, Holder RL, et al. Combining estimates of interest in prognostic modelling studies after multiple imputation: current practice and guidelines. BMC Med Res Methodol. 2009;9:57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Kutyla MJ, O’Connor S, Hourigan LF, et al. An Evidence-based approach towards targeted patient education to improve bowel Preparation for colonoscopy. J Clin Gastroenterol. 2020;54:707–13. [DOI] [PubMed] [Google Scholar]
- 16.Berger A, Cesbron-Métivier E, Bertrais S, et al. A predictive score of inadequate bowel Preparation based on a self-administered questionnaire: PREPA-CO. Clin Res Hepatol Gastroenterol. 2021;45:101693. [DOI] [PubMed] [Google Scholar]
- 17.Chen L, Ren G, Luo H, et al. Superiority of a Preparation-related model for predicting inadequate bowel Preparation in patients undergoing colonoscopy: A multicenter prospective study. J Gastroenterol Hepatol. 2022;37:2297–305. [DOI] [PubMed] [Google Scholar]
- 18.Malkin D, Cohen DL, Richter V, et al. A novel model to predict inadequate bowel Preparation prior to colonoscopy incorporating patients’ reactions to drinking the laxative. J Clin Med. 2023;12:7335. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Fang L, Bingbing W, Wang Q, et al. Development and validation of a nomogram predictive model and scoring tool for assessing the risk of inadequate bowel Preparation in colonoscopy patients over 40 years old: A retrospective observational study. J Clin Nurs. 2025;34:4398–414. [DOI] [PubMed] [Google Scholar]
- 20.Xu Y, Ou S, Yuan X, et al. Construction and validation of risk scoring model for inadequate bowel preparation in elderly patients undergoing colonoscopy. Ther Clin Risk Manage. 2025;21:885–94. [DOI] [PMC free article] [PubMed]
- 21.Zhao X, Pan Y, Hao J, et al. Development and validation of a novel scoring system based on a nomogram for predicting inadequate bowel Preparation. Clin Transl Oncol. 2024;26:2262–73. [DOI] [PubMed] [Google Scholar]
- 22.Demling L. Is crohn’s disease caused by antibiotics? Hepatogastroenterology. 1994;41:549–51. [PubMed] [Google Scholar]
- 23.Kurlander JE, Waljee AK, Menees SB, et al. Regression and random forest machine learning have limited performance in predicting bowel Preparation in veteran population. Dig Dis Sci. 2022;67:2827–41. [DOI] [PubMed] [Google Scholar]
- 24.Fuccio L, Frazzoni L, Spada C, et al. Factors that affect adequacy of colon cleansing for colonoscopy in hospitalized patients. Clin Gastroenterol Hepatol. 2021;19:339–48. e7. [DOI] [PubMed] [Google Scholar]
- 25.Hassan C, Fuccio L, Bruno M, et al. A predictive model identifies patients most likely to have inadequate bowel Preparation for colonoscopy. Clin Gastroenterol Hepatol. 2012;10:501–6. [DOI] [PubMed] [Google Scholar]
- 26.Xu M. Development and validation of a prediction model for inadequate bowel Preparation for colonoscopy in elderly patients. Shanxi Medical University; 2022.
- 27.Du Y, Neng Q, Li Y, et al. Gastrointestinal autonomic neuropathy exacerbates gut microbiota dysbiosis in adult patients with type 2 diabetes mellitus. Front Cell Infect Microbiol. 2021;11:804733. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Eshun-Wilson I, Siegfried N, Akena DH, et al. Antidepressants for depression in adults with HIV infection. Cochrane Database Syst Rev. 2018;1:CD008525. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Gillis A, Gondin AB, Kliewer A, et al. Low intrinsic efficacy for G protein activation can explain the improved side effect profiles of new opioid agonists. Sci Signal. 2020;13:eaaz3140. [DOI] [PubMed] [Google Scholar]
- 30.Dimitriadis K, Narkiewicz K, Leontsinis I, et al. Acute effects of electronic and tobacco cigarette smoking on sympathetic nerve activity and blood pressure in humans. Int J Environ Res Public Health. 2022;19:3237. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets analyzed during the current study are not publicly available due to ethical considerations, but are available from the corresponding author on reasonable request.