Predicting the Incidence of Pressure Ulcers in the Intensive Care Unit Using Machine Learning.
EGEMS (Washington, DC)
2019; 7 (1): 49
Background: Reducing hospital-acquired pressure ulcers (PUs) in intensive care units (ICUs) has emerged as an important quality metric for health systems internationally. Limited work has been done to characterize the profile of PUs in the ICU using observational data from the electronic health record (EHR). Consequently, there are limited EHR-based prognostic tools for determining a patient's risk of PU development, with most institutions relying on nurse-calculated risk scores such as the Braden score to identify high-risk patients.Methods and Results: Using EHR data from 50,851 admissions in a tertiary ICU (MIMIC-III), we show that the prevalence of PUs at stage 2 or above is 7.8 percent. For the 1,690 admissions where a PU was recorded on day 2 or beyond, we evaluated the prognostic value of the Braden score measured within the first 24 hours. A high-risk Braden score (<=12) had precision 0.09 and recall 0.50 for the future development of a PU. We trained a range of machine learning algorithms using demographic parameters, diagnosis codes, laboratory values and vitals available from the EHR within the first 24 hours. A weighted linear regression model showed precision 0.09 and recall 0.71 for future PU development. Classifier performance was not improved by integrating Braden score elements into the model.Conclusion: We demonstrate that an EHR-based model can outperform the Braden score as a screening tool for PUs. This may be a useful tool for automatic risk stratification early in an admission, helping to guide quality protocols in the ICU, including the allocation and timing of prophylactic interventions.
View details for DOI 10.5334/egems.307
View details for PubMedID 31534981
Factors Associated With Acute Pain Estimation, Postoperative Pain Resolution, Opioid Cessation, and Recovery: Secondary Analysis of a Randomized Clinical Trial.
JAMA network open
2019; 2 (3): e190168
Importance: Acute postoperative pain is associated with the development of persistent postsurgical pain, but it is unclear which aspect is most estimable.Objective: To identify patient clusters based on acute pain trajectories, preoperative psychosocial characteristics associated with the high-risk cluster, and the best acute pain predictor of remote outcomes.Design, Setting, and Participants: A secondary analysis of the Stanford Accelerated Recovery Trial randomized, double-blind clinical trial was conducted at a single-center, tertiary, referral teaching hospital. A total of 422 participants scheduled for thoracotomy, video-assisted thoracoscopic surgery, total hip replacement, total knee replacement, mastectomy, breast lumpectomy, hand surgery, carpal tunnel surgery, knee arthroscopy, shoulder arthroplasty, or shoulder arthroscopy were enrolled between May 25, 2010, and July 25, 2014. Data analysis was performed from January 1 to August 1, 2018.Interventions: Patients were randomized to receive gabapentin (1200 mg, preoperatively, and 600 mg, 3 times a day postoperatively) or active placebo (lorazepam, 0.5 mg preoperatively, inactive placebo postoperatively) for 72 hours.Main Outcomes and Measures: A modified Brief Pain Inventory prospectively captured 3 surgical site pain outcomes: average pain and worst pain intensity over the past 24 hours, and current pain intensity. Within each category, acute pain trajectories (first 10 postoperative pain scores) were compared using a k-means clustering algorithm. Fifteen descriptors of acute pain were compared as predictors of remote postoperative pain resolution, opioid cessation, and full recovery.Results: Of the 422 patients enrolled, 371 patients (≤10% missing pain scores) were included in the analysis. Of these, 146 (39.4%) were men; mean (SD) age was 56.67 (11.70) years. Two clusters were identified within each trajectory category. The high pain cluster of the average pain trajectory significantly predicted prolonged pain (hazard ratio [HR], 0.63; 95% CI, 0.50-0.80; P<.001) and delayed opioid cessation (HR, 0.52; 95% CI, 0.41-0.67; P<.001) but was not a predictor of time to recovery in Cox proportional hazards regression (HR, 0.89; 95% CI, 0.69-1.14; P=.89). Preoperative risk factors for categorization to the high average pain cluster included female sex (adjusted relative risk [ARR], 1.36; 95% CI, 1.08-1.70; P=.008), elevated preoperative pain (ARR, 1.11; 95% CI, 1.07-1.15; P<.001), a history of alcohol or drug abuse treatment (ARR,1.90; 95% CI, 1.42-2.53; P<.001), and receiving active placebo (ARR, 1.27; 95% CI, 1.03-1.56; P=.03). Worst pain reported on postoperative day 10 was the best predictor of time to pain resolution (HR, 0.83; 95% CI, 0.78-0.87; P<.001), opioid cessation (HR, 0.84; 95% CI, 0.80-0.89; P<.001), and complete surgical recovery (HR, 0.91; 95% CI, 0.86-0.96; P<.001).Conclusions and Relevance: This study has shown a possible uniform predictor of remote postoperative pain, opioid use, and recovery that can be easily assessed. Future work is needed to replicate these findings.Trial Registration: ClinicalTrials.gov Identifier: NCT01067144.
View details for PubMedID 30821824
- Factors Associated With Acute Pain Estimation, Postoperative Pain Resolution, Opioid Cessation, and Recovery Secondary Analysis of a Randomized Clinical Trial JAMA NETWORK OPEN 2019; 2 (3)