Dataset Information

Predictors of 30-Day Mortality Among Dutch Patients Undergoing Colorectal Cancer Surgery, 2011-2016.

ABSTRACT:

Importance

Quality improvement programs for colorectal cancer surgery have been introduced with benchmarking based on quality indicators, such as mortality. Detailed (pre)operative characteristics may offer relevant information for proper case-mix correction.

Objective

To investigate the added value of machine learning to predict quality indicators for colorectal cancer surgery and identify previously unrecognized predictors of 30-day mortality based on a large, nationwide colorectal cancer registry that collected extensive data on comorbidities.

Design, setting, and participants

All patients who underwent resection for primary colorectal cancer registered in the Dutch ColoRectal Audit between January 1, 2011, and December 31, 2016, were included. Multiple machine learning models (multivariable logistic regression, elastic net regression, support vector machine, random forest, and gradient boosting) were made to predict quality indicators. Model performance was compared with conventionally used scores. Risk factors were identified by logistic regression analyses and Shapley additive explanations (ie, SHAP values). Statistical analysis was performed between March 1 and September 30, 2020.

Main outcomes and measures

The primary outcome of this cohort study was 30-day mortality. Prediction models were trained on a training set by performing 5-fold cross-validation, and outcomes were measured by the area under the receiver operating characteristic curve on the test set. Machine learning was further used to identify risk factors, measured by odds ratios and SHAP values.

Results

This cohort study included 62 501 records, most patients were male (35 116 [56.2%]), were aged 61 to 80 years (41 560 [66.5%]), and had an American Society of Anesthesiology score of II (35 679 [57.1%]). A 30-day mortality rate of 2.7% (n = 1693) was found. The area under the curve of the best machine learning model for 30-day mortality (0.82; 95% CI, 0.79-0.85) was significantly higher than the American Society of Anesthesiology score (0.74; 95% CI, 0.71-0.77; P < .001), Charlson Comorbidity Index (0.66; 95% CI, 0.63-0.70; P < .001), and preoperative score to predict postoperative mortality (0.73; 95% CI, 0.70-0.77; P < .001). Hypertension, myocardial infarction, chronic obstructive pulmonary disease, and asthma were comorbidities with a high risk for increased mortality. Machine learning identified specific risk factors for a complicated course, intensive care unit admission, prolonged hospital stay, and readmission. Laparoscopic surgery was associated with a decreased risk for all adverse outcomes.

Conclusions and relevance

This study found that machine learning methods outperformed conventional scores to predict 30-day mortality after colorectal cancer surgery, identified specific patient groups at risk for adverse outcomes, and provided directions to optimize benchmarking in clinical audits.

SUBMITTER: van den Bosch T

PROVIDER: S-EPMC8076964 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Importance:Although wait times for hip fracture surgery have been linked to mortality and are being used as quality-of-care indicators worldwide, controversy exists about the duration of the wait that leads to complications. Objective:To use population-based wait-time data to identify the optimal time window in which to conduct hip fracture surgery before the risk of complications increases. Design, Setting, and Participants:Population-based, retrospective cohort study of adults undergoing hip fracture surgery between April 1, 2009, and March 31, 2014, at 72 hospitals in Ontario, Canada. Risk-adjusted restricted cubic splines modeled the probability of each complication according to wait time. The inflection point (in hours) when complications began to increase was used to define early and delayed surgery. To evaluate the robustness of this definition, outcomes among propensity-score matched early and delayed surgical patients were compared using percent absolute risk differences (RDs, with 95% CIs). Exposure:Time elapsed from hospital arrival to surgery (in hours). Main Outcomes and Measures:Mortality within 30 days. Secondary outcomes included a composite of mortality or other medical complications (myocardial infarction, deep vein thrombosis, pulmonary embolism, and pneumonia). Results:Among 42 230 patients with hip fracture (mean [SD] age, 80.1 years [10.7], 70.5% women) who met study entry criteria, overall mortality at 30 days was 7.0%. The risk of complications increased when wait times were greater than 24 hours, irrespective of the complication considered. Compared with 13 731 propensity-score matched patients who received surgery earlier, 13 731 patients who received surgery after 24 hours had a significantly higher risk of 30-day mortality (898 [6.5%] vs 790 [5.8%]; % absolute RD, 0.79; 95% CI, 0.23-1.35) and the composite outcome (1680 [12.2%]) vs 1383 [10.1%]; % absolute RD, 2.16; 95% CI, 1.43-2.89). Conclusions and Relevance:Among adults undergoing hip fracture surgery, increased wait time was associated with a greater risk of 30-day mortality and other complications. A wait time of 24 hours may represent a threshold defining higher risk.

Project description:Postoperative mortality after colorectal cancer surgery varies across hospitals and countries. The aim of this study was to test the Association of Coloproctologists of Great Britain and Ireland (ACPGBI) models as predictors of 30-day mortality in an Australian cohort. Data from patients who underwent surgery in six hospitals between 1996 and 2015 (CRC data set) were reviewed to test ACPGBI models, and patients from 79 hospitals in the Bi-National Colorectal Cancer Audit between 2007 and 2016 (BCCA data set) were analysed to validate model performance. Recalibrated models based on ACPGBI risk models were developed, tested and validated on a data set of Australasian patients. Of 18 752 patients observed during the study, 6727 (CRC data set) and 3814 (BCCA data set) were analysed. The 30-day mortality rate was 1·1 and 3·5 per cent in the CRC and BCCA data sets respectively. Both the original and revised ACPGBI models overestimated 30-day mortality for the CRC data set (observed to expected (O/E) ratio 0·17 and 0·21 respectively). Their ability to correctly predict mortality risk was poor (P < 0·001, Hosmer-Lemeshow test); however, the area under the curve for both models was 0·88 (95 per cent c.i. 0·85 to 0·92) showing good discriminatory power to classify 30-day mortality. The recalibrated original model performed well for calibration and discrimination, whereas the recalibrated revised model performed well for discrimination but not for calibration. Risk prediction was good for both recalibrated models. On external validation using the BCCA data set, the recalibrated models underestimated mortality risk (O/E ratio 3·06 and 2·98 respectively), whereas both original and revised ACPGBI models overestimated the risk (O/E ratio 0·48 and 0·69). All models showed similar good discrimination. The original and revised ACPGBI models overpredicted risk of 30-day mortality. The new Australasian calibrated ACPGBI model needs to be tested further in clinical practice.

Project description:OBJECTIVE:Cardiac surgery ICU characteristics and clinician staffing patterns have not been well characterized. We sought to describe Pennsylvania cardiac ICUs and to determine whether ICU characteristics are associated with mortality in the 30 days after cardiac surgery. DESIGN:From 2012 to 2013, we conducted a survey of cardiac surgery ICUs in Pennsylvania to assess ICU structure, care practices, and clinician staffing patterns. ICU data were linked to an administrative database of cardiac surgery patient discharges. We used logistic regression to measure the association between ICU variables and death in 30 days. SETTING:Cardiac surgery ICUs in Pennsylvania. PATIENTS:Patients having coronary artery bypass grafting and/or cardiac valve repair or replacement from 2009 to 2011. INTERVENTIONS:None. MEASUREMENTS AND MAIN RESULTS:Of the 57 cardiac surgical ICUs in Pennsylvania, 43 (75.4%) responded to the facility survey. Rounds included respiratory therapists in 26 of 43 (60.5%) and pharmacists in 23 of 43 (53.5%). Eleven of 41 (26.8%) reported that at least 2/3 of their nurses had a bachelor's degree in nursing. Advanced practice providers were present in most of the ICUs (37/43; 86.0%) but residents (8/42; 18.6%) and fellows (7/43; 16.3%) were not. Daytime intensivists were present in 21 of 43 (48.8%) responding ICUs; eight of 43 (18.6%) had nighttime intensivists. Among 29,449 patients, there was no relationship between mortality and nurse ICU experience, presence of any intensivist, or absence of residents after risk adjustment. To exclude patients who may have undergone transcatheter aortic valve replacement, we conducted a subgroup analysis of patients undergoing only coronary artery bypass grafting, and results were similar. CONCLUSIONS:Pennsylvania cardiac surgery ICUs have variable structures, care practices, and clinician staffing, although none of these are statistically significantly associated with mortality in the 30 days following surgery after adjustment.

Project description:PurposeThe 30-day mortality rate after hip fracture surgery has been considered as an indirect indicator of the quality of care. The aim of this work is to analyse preoperative and postoperative factors potentially related to early 30-day mortality in patients over 65 undergoing hip fracture surgery.MethodsProspective cohort study including all consecutive primary hip fracture patients over 65 admitted to Trauma and Orthopaedics department from January 1, 2018 to December 31, 2019. Bed-ridden, non- surgically treated patients, and high energy trauma or tumoral aetiology fractures were excluded. A total of 943 patients were eligible (attrition rate: 2.1%). Follow-up included 30-days after discharge. We noted the 30-day mortality after hip fracture surgery, analysing 130 potentially related variables including biodemographic, fracture-related, preoperative, and postoperative clinical factors. Qualitative variables were assessed by χ2, and quantitative variables by non-parametric tests. Odds ratio determined by binary logistic regression. We selected preventable candidate variables for multivariate risk assessment by logistic regression.ResultsA total of 923 patients were enrolled (mean age 86.22±6.8, 72.9% women). The 30-day mortality rate was 6.0%. We noted significant increased mortality on men (OR = 2.381[1.371-4.136], p = 0.002), ageing patients (ORyear = 1.073[1.025-1.122], p = 0.002), and longer time to surgery (ORday = 1.183[1.039-1146], p<0.001), on other 20 preoperative clinical variables, like lymphopenia (lymphocyte count <103/μl, OR = 1.842[1.063-3.191], p = 0.029), hypoalbuminemia (≤3.5g/dl, OR = 2.474[1.316-4.643], p = 0.005), and oral anticoagulant intake (OR = 2.499[1.415-4.415], p = 0.002), and on 25 postoperative clinical variables, like arrhythmia (OR = 13.937[6.263-31.017], p<0.001), respiratory insufficiency (OR = 7.002[3.947-12.419], p<0.001), hyperkalaemia (OR = 10.378[3.909-27.555], p<0.001), nutritional supply requirement (OR = 3.576[1.894-6.752], p = 0.021), or early arthroplasty dislocation (OR = 6.557[1.206-35.640], p = 0.029). We developed a predictive model for early mortality after hip fracture surgery based on postoperative factors with 96.0% sensitivity and 60.7% specificity (AUC = 0.863).ConclusionWe revealed that not only preoperative, but also postoperative factors have a great impact after hip fracture surgery. The influence of post-operative factors on 30-day mortality has a logical basis, albeit so far they have not been identified or quantified before. Our results provide an advantageous picture of the 30-day mortality after hip fracture surgery.