Citations

Abstract In the U.S. inpatient payment system, the Diagnosis-Related Group (DRG) is pivotal, but its assignment process is inefficient. The study introduces DRG-LLaMA, an advanced large language model (LLM) fine-tuned on clinical notes to enhance DRGs assignment. Utilizing LLaMA as the foundational model and optimizing it through Low-Rank Adaptation (LoRA) on 236,192 MIMIC-IV discharge summaries, our DRG-LLaMA-7B model exhibited a noteworthy macro-averaged F1 score of 0.327, a top-1 prediction accuracy of 52.0%, and a macro-averaged Area Under the Curve (AUC) of 0.986, with a maximum input token length of 512. This model surpassed the performance of prior leading models in DRG prediction, showing a relative improvement of 40.3% and 35.7% in macro-averaged F1 score compared to ClinicalBERT and CAML, respectively. Applied to base DRG and complication or comorbidity (CC)/major complication or comorbidity (MCC) prediction, DRG-LLaMA achieved a top-1 prediction accuracy of 67.8% and 67.5%, respectively. Additionally, our findings indicate that DRG-LLaMA's performance correlates with increased model parameters and input context lengths.

Background: Treating hyperglycemia promotes better outcomes among inpatients. Knowledge deficits about management of inpatient diabetes are prevalent among resident physicians, which may affect the care of a substantial number of these patients. Methods: A computer-based training (CBT) curriculum on inpatient diabetes and hyperglycemia was developed and implemented for use by resident physicians and focuses on several aspects of the management of inpatient diabetes and hyperglycemia: (1) review of importance of inpatient glucose control, (2) overview of institution-specific data, (3) triaging and initial admission actions for diabetes or hyperglycemia, (4) overview of pharmacologic management, (5) insulin-dosing calculations and ordering simulations, (6) review of existing policies and procedures, and (7) discharge planning. The curriculum was first provided as a series of lectures, then formatted and placed on the institutional intranet as a CBT program. Results: Residents began using the inpatient CBT in September 2008. By August 2009, a total of 29 residents had participated in CBT: 8 in family medicine, 12 in internal medicine, and 9 in general surgery. Most of the 29 residents confirmed that module content met stated objectives, considered the information valuable to their inpatient practices, and believed that the quality of the online modules met expectations. The majority reported that the modules took just the right amount of time to complete (typically 30 min each). Conclusions: Improvement in inpatient diabetes care requires continuous educational efforts. The CBT format and curriculum content were well accepted by the resident physicians. Ongoing assessment must determine whether resident practice patterns are influenced by such training.

Background: Hypoglycemia is often cited as a barrier to achieving inpatient glycemic targets. We sought to characterize hypoglycemic events in our institution by work-shift cycle and by specific treatment area. Methods: Capillary (bedside) and blood (laboratory) glucose values of <70 mg/dl for patients with either a known diagnosis of diabetes or with evidence of hyperglycemia were abstracted from our laboratory database for hospitalizations between October 1, 2007, and February 3, 2008. Hypoglycemic events were analyzed by 12 h nursing work-shift cycles (day shift, 07:00 to 18:59; night shift, 19:00 to 06:59) and by the six medical, surgical, and intensive care areas in the hospital (designated areas 1 to 6). Results: We identified 206 individual patients with either diabetes or hyperglycemia (mean age, 67 years; 56% men; 83% white) who had 423 hypoglycemic events. There were 78% more hypoglycemic events during the night shift ( n = 271 events in 128 individual patients) than during the day shift ( n = 152 events in 96 individual patients). Most of the night-shift hypoglycemic measurements were detected between 04:00 and 04:59 or 06:00 and 06:59. The mean hypoglycemic level was comparable between shifts ( p = .79) and across the six inpatient areas. The number of hypoglycemic events per person increased with lengths of hospital stay >5 days. The prevalence of hypoglycemia varied across patient care areas within the hospital, with most (28%) detected in one area of the hospital. Conclusion: There are temporal and geographic patterns in the occurrence of hypoglycemia among patients with diabetes or hyperglycemia in our hospital. Further study should focus on the reasons underlying these variations so that specific interventions can address the risk of hypoglycemia during peak times and places.