Characterizing the limitations of using diagnosis codes in the context of machine learning for healthcare.

BACKGROUND: Diagnostic codes are commonly used as inputs for clinical prediction models, to create labels for prediction tasks, and to identify cohorts for multicenter network studies. However, the coverage rates of diagnostic codes and their variability across institutions are underexplored. The primary objective was to describe lab- and diagnosis-based labels for 7 selected outcomes at three institutions. Secondary objectives were to describe agreement, sensitivity, and specificity of diagnosis-based labels against lab-based labels. METHODS: This study included three cohorts: SickKids from The Hospital for Sick Children, and StanfordPeds and StanfordAdults from Stanford Medicine. We included seven clinical outcomes with lab-based definitions: acute kidney injury, hyperkalemia, hypoglycemia, hyponatremia, anemia, neutropenia and thrombocytopenia. For each outcome, we created four lab-based labels (abnormal, mild, moderate and severe) based on test result and one diagnosis-based label. Proportion of admissions with a positive label were presented for each outcome stratified by cohort. Using lab-based labels as the gold standard, agreement using Cohen's Kappa, sensitivity and specificity were calculated for each lab-based severity level. RESULTS: The number of admissions included were: SickKids (n = 59,298), StanfordPeds (n = 24,639) and StanfordAdults (n = 159,985). The proportion of admissions with a positive diagnosis-based label was significantly higher for StanfordPeds compared to SickKids across all outcomes, with odds ratio (99.9% confidence interval) for abnormal diagnosis-based label ranging from 2.2 (1.7-2.7) for neutropenia to 18.4 (10.1-33.4) for hyperkalemia. Lab-based labels were more similar by institution. When using lab-based labels as the gold standard, Cohen's Kappa and sensitivity were lower at SickKids for all severity levels compared to StanfordPeds. CONCLUSIONS: Across multiple outcomes, diagnosis codes were consistently different between the two pediatric institutions. This difference was not explained by differences in test results. These results may have implications for machine learning model development and deployment.

Development and validation of the SickKids Enterprise-wide Data in Azure Repository (SEDAR).

Objectives: To describe the processes developed by The Hospital for Sick Children (SickKids) to enable utilization of electronic health record (EHR) data by creating sequentially transformed schemas for use across multiple user types. Methods: We used Microsoft Azure as the cloud service provider and named this effort the SickKids Enterprise-wide Data in Azure Repository (SEDAR). Epic Clarity data from on-premises was copied to a virtual network in Microsoft Azure. Three sequential schemas were developed. The Filtered Schema added a filter to retain only SickKids and valid patients. The Curated Schema created a data structure that was easier to navigate and query. Each table contained a logical unit such as patients, hospital encounters or laboratory tests. Data validation of randomly sampled observations in the Curated Schema was performed. The SK-OMOP Schema was designed to facilitate research and machine learning. Two individuals mapped medical elements to standard Observational Medical Outcomes Partnership (OMOP) concepts. Results: A copy of Clarity data was transferred to Microsoft Azure and updated each night using log shipping. The Filtered Schema and Curated Schema were implemented as stored procedures and executed each night with incremental updates or full loads. Data validation required up to 16 iterations for each Curated Schema table. OMOP concept mapping achieved at least 80 % coverage for each SK-OMOP table. Conclusions: We described our experience in creating three sequential schemas to address different EHR data access requirements. Future work should consider replicating this approach at other institutions to determine whether approaches are generalizable.

Self-supervised machine learning using adult inpatient data produces effective models for pediatric clinical prediction tasks.

OBJECTIVE: Development of electronic health records (EHR)-based machine learning models for pediatric inpatients is challenged by limited training data. Self-supervised learning using adult data may be a promising approach to creating robust pediatric prediction models. The primary objective was to determine whether a self-supervised model trained in adult inpatients was noninferior to logistic regression models trained in pediatric inpatients, for pediatric inpatient clinical prediction tasks. MATERIALS AND METHODS: This retrospective cohort study used EHR data and included patients with at least one admission to an inpatient unit. One admission per patient was randomly selected. Adult inpatients were 18 years or older while pediatric inpatients were more than 28 days and less than 18 years. Admissions were temporally split into training (January 1, 2008 to December 31, 2019), validation (January 1, 2020 to December 31, 2020), and test (January 1, 2021 to August 1, 2022) sets. Primary comparison was a self-supervised model trained in adult inpatients versus count-based logistic regression models trained in pediatric inpatients. Primary outcome was mean area-under-the-receiver-operating-characteristic-curve (AUROC) for 11 distinct clinical outcomes. Models were evaluated in pediatric inpatients. RESULTS: When evaluated in pediatric inpatients, mean AUROC of self-supervised model trained in adult inpatients (0.902) was noninferior to count-based logistic regression models trained in pediatric inpatients (0.868) (mean difference = 0.034, 95% CI=0.014-0.057; P < .001 for noninferiority and P = .006 for superiority). CONCLUSIONS: Self-supervised learning in adult inpatients was noninferior to logistic regression models trained in pediatric inpatients. This finding suggests transferability of self-supervised models trained in adult patients to pediatric patients, without requiring costly model retraining.

EHR foundation models improve robustness in the presence of temporal distribution shift.

Temporal distribution shift negatively impacts the performance of clinical prediction models over time. Pretraining foundation models using self-supervised learning on electronic health records (EHR) may be effective in acquiring informative global patterns that can improve the robustness of task-specific models. The objective was to evaluate the utility of EHR foundation models in improving the in-distribution (ID) and out-of-distribution (OOD) performance of clinical prediction models. Transformer- and gated recurrent unit-based foundation models were pretrained on EHR of up to 1.8 M patients (382 M coded events) collected within pre-determined year groups (e.g., 2009-2012) and were subsequently used to construct patient representations for patients admitted to inpatient units. These representations were used to train logistic regression models to predict hospital mortality, long length of stay, 30-day readmission, and ICU admission. We compared our EHR foundation models with baseline logistic regression models learned on count-based representations (count-LR) in ID and OOD year groups. Performance was measured using area-under-the-receiver-operating-characteristic curve (AUROC), area-under-the-precision-recall curve, and absolute calibration error. Both transformer and recurrent-based foundation models generally showed better ID and OOD discrimination relative to count-LR and often exhibited less decay in tasks where there is observable degradation of discrimination performance (average AUROC decay of 3% for transformer-based foundation model vs. 7% for count-LR after 5-9 years). In addition, the performance and robustness of transformer-based foundation models continued to improve as pretraining set size increased. These results suggest that pretraining EHR foundation models at scale is a useful approach for developing clinical prediction models that perform well in the presence of temporal distribution shift.

Evaluation of Feature Selection Methods for Preserving Machine Learning Performance in the Presence of Temporal Dataset Shift in Clinical Medicine.

BACKGROUND: Temporal dataset shift can cause degradation in model performance as discrepancies between training and deployment data grow over time. The primary objective was to determine whether parsimonious models produced by specific feature selection methods are more robust to temporal dataset shift as measured by out-of-distribution (OOD) performance, while maintaining in-distribution (ID) performance. METHODS: Our dataset consisted of intensive care unit patients from MIMIC-IV categorized by year groups (2008-2010, 2011-2013, 2014-2016, and 2017-2019). We trained baseline models using L2-regularized logistic regression on 2008-2010 to predict in-hospital mortality, long length of stay (LOS), sepsis, and invasive ventilation in all year groups. We evaluated three feature selection methods: L1-regularized logistic regression (L1), Remove and Retrain (ROAR), and causal feature selection. We assessed whether a feature selection method could maintain ID performance (2008-2010) and improve OOD performance (2017-2019). We also assessed whether parsimonious models retrained on OOD data performed as well as oracle models trained on all features in the OOD year group. RESULTS: The baseline model showed significantly worse OOD performance with the long LOS and sepsis tasks when compared with the ID performance. L1 and ROAR retained 3.7 to 12.6% of all features, whereas causal feature selection generally retained fewer features. Models produced by L1 and ROAR exhibited similar ID and OOD performance as the baseline models. The retraining of these models on 2017-2019 data using features selected from training on 2008-2010 data generally reached parity with oracle models trained directly on 2017-2019 data using all available features. Causal feature selection led to heterogeneous results with the superset maintaining ID performance while improving OOD calibration only on the long LOS task. CONCLUSIONS: While model retraining can mitigate the impact of temporal dataset shift on parsimonious models produced by L1 and ROAR, new methods are required to proactively improve temporal robustness.

A 25-year experience with hemicorporectomy for terminal pelvic osteomyelitis.

BACKGROUND: Hemicorporectomy involves amputation of the pelvis and lower extremities by disarticulation through the lumbar spine with concomitant transection of the aorta, inferior vena cava, and spinal cord. In addition, conduits are constructed for diversion of both the urinary and fecal streams. Of 57 cases reported in the literature, limited experience exists with hemicorporectomy for terminal pelvic osteomyelitis, with only 11 cases described. Furthermore, there is little information available regarding perioperative mortality and long-term survival. This article describes the largest reported series of hemicorporectomies performed for terminal pelvic osteomyelitis. METHODS: A retrospective review of the medical records for nine patients who underwent hemicorporectomy at the authors' institution was conducted followed by interviews with all surviving patients. RESULTS: At follow-up, four patients were alive and five patients were dead. For all patients, the average survival after hemicorporectomy was 11.0 years (range, 1.7 to 22.0 years). There was no perioperative mortality within 30 days of surgery. None of the surviving patients suffered from recurrent decubitus ulcers. CONCLUSIONS: Including this clinical series, a total of 66 hemicorporectomies have now been reported in the literature. Twenty cases were performed for terminal pelvic osteomyelitis with no mortality within 30 days of surgery, and 53.3 percent of patients were alive and well at long-term follow-up. Given the low perioperative mortality along with the ability of patients to achieve long-term survival following this operation, hemicorporectomy should be offered to appropriate patients suffering from terminal pelvic osteomyelitis.

The individualized component face lift: developing a systematic approach to facial rejuvenation.

BACKGROUND: Accurate preoperative planning combined with facial fat compartment augmentation can improve precision and balance in facial rejuvenation techniques. Understanding the concept of "facial shaping" with respect to symmetry and soft-tissue (fat) distribution preoperatively is critical to optimizing aesthetic outcomes in various face lift techniques. METHODS: A review of 822 consecutive face lifts performed from January of 1994 to June of 2007 by a single surgeon (R.J.R.) was conducted. From this database, randomly selected cohorts of 50 preoperative and postoperative photographs were critically analyzed by three plastic surgeons exclusive of the senior surgeon (R.J.R.). Three facial parameters were compared on each facial side: facial height, degree of malar deflation, and orbit size. Long-term improvement was evaluated to delineate factors contributing to success in creating an aesthetically balanced facial shape. RESULTS: Asymmetry between the two facial sides was noted in every patient preoperatively with respect to the three study parameters and was improved postoperatively. There was no statistically significant interobserver bias in the evaluations (p < 0.005). Facial asymmetry dictated differential treatment of the superficial musculoaponeurotic system (SMAS) tissue between facial sides to achieve the desired youthful facial shape. The angle (vector) and extent of SMAS-stacking varied depending on the preoperative analysis. Similarly, the selection of SMAS-ectomy versus SMAS-stacking depended on the degree of malar deflation and resultant cheek fullness. CONCLUSIONS: Proper preoperative analysis for evaluating facial shape should address (1) facial height, (2) facial width, and (3) overall distribution/location of facial fullness. This method of evaluating facial shape and symmetry is simple and reproducible, and can aid in formulating a comprehensive treatment plan.

Hemicorporectomy: back to front.

Hemicorporectomy involves amputation of the pelvis and lower extremities by disarticulation through the lumbar spine with concomitant transection of the aorta, inferior vena cava, and spinal cord, as well as creation of conduits for diversion of the urinary and fecal streams. A review of the literature reveals that the surgical technique has been relatively unchanged since 1960. The standard anterior to posterior approach is associated with significant blood loss and morbidity, likely contributing to lengthy hospital stay. Herein, we describe our back-to-front approach to hemicorporectomy, involving early division of the vertebral structures and spinal cord, pre-empting engorgement of Batson's plexus, thus minimizing blood loss. In addition, this approach greatly improves exposure of the pelvic vessels, allowing for a technically less challenging and safer procedure.

Soft-tissue injuries of the fingertip: methods of evaluation and treatment. An algorithmic approach.

LEARNING OBJECTIVES: After studying this article, the participant should be able to: 1. Understand the anatomy of the fingertip. 2. Describe the methods of evaluating fingertip injuries. 3. Discuss reconstructive options for various tip injuries. SUMMARY: The fingertip is the most commonly injured part of the hand, and therefore fingertip injuries are among the most frequent injuries that plastic surgeons are asked to treat. Although microsurgical techniques have enabled replantation of even very distal tip amputations, it is relatively uncommon that a distal tip injury will be appropriate for replantation. In the event that replantation is not pursued, options for distal tip soft-tissue reconstruction must be considered. This review presents a straightforward method for evaluating fingertip injuries and provides an algorithm for fingertip reconstruction.

The anatomy of the corrugator supercilii muscle: part II. Supraorbital nerve branching patterns.

BACKGROUND: This article focuses on delineation of supraorbital nerve branching patterns relative to the corrugator muscle fibers and identifies four branching patterns that help improve understanding of the local anatomy. METHODS: Twenty-five fresh cadaver heads (50 corrugator supercilii muscles and 50 supraorbital nerves) were dissected and the corrugator supercilii muscles isolated. After corrugator supercilii muscle measurement points were recorded for part I of the study, the supraorbital nerve branches were then traced from their emergence points from the orbit and dissected out to the defined topographical boundaries of the muscle. Nerve branching patterns relative to the muscle fibers were analyzed, and a classification system for branching patterns relative to the muscle was created. RESULTS: Four types of supraorbital nerve branching patterns were found. In type I (40 percent), only the deep supraorbital nerve division sent branches that coursed directly along the undersurface of the muscle. In type II (34 percent), branches emerging directly from the superficial supraorbital nerve were found in addition to the branches from the deep division. Type III (4 percent) included discrete branches from the superficial division, but none from the deep division. In type IV (22 percent), significant branching began more cephalad relative to the muscle and, therefore, displayed no specific relation to the muscle fibers. CONCLUSIONS: Contrary to previous reports, both the deep and superficial divisions of the supraorbital nerve are intimately associated with corrugator supercilii muscle fibers. Four supraorbital nerve branching patterns from these divisions were found. Potential sites of supraorbital nerve compression were identified. This more detailed anatomical information may improve the safety and accuracy of performing complete corrugator supercilii muscle resection.

Anatomy of the corrugator supercilii muscle: part I. Corrugator topography.

BACKGROUND: Complete corrugator supercilii muscle resection is important for the surgical treatment of migraine headaches and may help prevent postoperative abnormalities in surgical forehead rejuvenation. Specific topographic analysis of corrugator supercilii muscle dimensions and its detailed association with the supraorbital nerve branching patterns has not been thoroughly delineated. Part I of this two-part study aims to define corrugator supercilii muscle topography with respect to external bony landmarks. METHODS: Twenty-five fresh cadaver heads (50 corrugator supercilii muscles and 50 supraorbital nerves) were dissected to isolate the corrugator supercilii muscle from surrounding muscles. Standardized measurements of corrugator supercilii muscle dimensions were taken with respect to the nasion and lateral orbital rim. RESULTS: Relative to the nasion, the most medial origin of the corrugator supercilii muscle was found at 2.9 +/- 1.0 mm; the most lateral origin point, 14.0 +/- 2.8 mm. The lateralmost insertion of the corrugator supercilii muscle measured 43.3 +/- 2.9 mm from the nasion or 7.6 +/- 2.7 mm medial to the lateral orbital rim. The most cephalic extent (apex) of the muscle was located 32.6 +/- 3.1 mm cephalad to the nasion-lateral orbital rim plane and 18.0 +/- 3.7 mm medial to the lateral orbital rim. There were no statistical differences noted between the right and left sides. CONCLUSIONS: The dimensions of the corrugator supercilii muscle are more extensive than previously described and can be easily delineated using fixed bony landmarks. These data may prove beneficial in performing safe, complete, and symmetric corrugator supercilii muscle resection for forehead rejuvenation and for effective decompression of the supraorbital nerve and supratrochlear nerve branches in the surgical treatment of migraine headaches.