Assessing the potential of GPT-4 to perpetuate racial and gender biases in health care: a model evaluation study.

BACKGROUND: Large language models (LLMs) such as GPT-4 hold great promise as transformative tools in health care, ranging from automating administrative tasks to augmenting clinical decision making. However, these models also pose a danger of perpetuating biases and delivering incorrect medical diagnoses, which can have a direct, harmful impact on medical care. We aimed to assess whether GPT-4 encodes racial and gender biases that impact its use in health care. METHODS: Using the Azure OpenAI application interface, this model evaluation study tested whether GPT-4 encodes racial and gender biases and examined the impact of such biases on four potential applications of LLMs in the clinical domain-namely, medical education, diagnostic reasoning, clinical plan generation, and subjective patient assessment. We conducted experiments with prompts designed to resemble typical use of GPT-4 within clinical and medical education applications. We used clinical vignettes from NEJM Healer and from published research on implicit bias in health care. GPT-4 estimates of the demographic distribution of medical conditions were compared with true US prevalence estimates. Differential diagnosis and treatment planning were evaluated across demographic groups using standard statistical tests for significance between groups. FINDINGS: We found that GPT-4 did not appropriately model the demographic diversity of medical conditions, consistently producing clinical vignettes that stereotype demographic presentations. The differential diagnoses created by GPT-4 for standardised clinical vignettes were more likely to include diagnoses that stereotype certain races, ethnicities, and genders. Assessment and plans created by the model showed significant association between demographic attributes and recommendations for more expensive procedures as well as differences in patient perception. INTERPRETATION: Our findings highlight the urgent need for comprehensive and transparent bias assessments of LLM tools such as GPT-4 for intended use cases before they are integrated into clinical care. We discuss the potential sources of these biases and potential mitigation strategies before clinical implementation. FUNDING: Priscilla Chan and Mark Zuckerberg.

Understanding Clinical Trial Reports: Extracting Medical Entities and Their Relations.

The best evidence concerning comparative treatment effectiveness comes from clinical trials, the results of which are reported in unstructured articles. Medical experts must manually extract information from articles to inform decision-making, which is time-consuming and expensive. Here we consider the end-to-end task of both (a) extracting treatments and outcomes from full-text articles describing clinical trials (entity identification) and, (b) inferring the reported results for the former with respect to the latter (relation extraction). We introduce new data for this task, and evaluate models that have recently achieved state-of-the-art results on similar tasks in Natural Language Processing. We then propose a new method motivated by how trial results are typically presented that outperforms these purely data-driven baselines. Finally, we run a fielded evaluation of the model with a non-profit seeking to identify existing drugs that might be re-purposed for cancer, showing the potential utility of end-to-end evidence extraction systems.

Representation Learning Approaches to Detect False Arrhythmia Alarms from ECG Dynamics.

The high rate of intensive care unit false arrhythmia alarms can lead to disruption of care and slow response time due to desensitization of clinical staff. We study the use of machine learning models to detect false ventricular tachycardia (v-tach) alarms using ECG waveform recordings. We propose using a Supervised Denoising Autoencoder (SDAE) to detect false alarms using a low-dimensional representation of ECG dynamics learned by minimizing a combined reconstruction and classification loss. We evaluate our algorithms on the PhysioNet Challenge 2015 dataset, containing over 500 records (over 300 training and 200 testing) with v-tach alarms. Our results indicate that using the SDAE on Fast Fourier Transformed (FFT) ECG at a beat-by-beat level outperforms several competitive baselines on the task of v-tach false alarm classification. We show that it is important to exploit the underlying known physiological structure using beat-by-beat frequency distribution from multiple cardiac cycles of the ECG waveforms to obtain competitive results and improve over previous entries from the 2015 PhysioNet Challenge.

Bedtime Use of Technology and Associated Sleep Problems in Children.

Children comprise one of the largest consumer groups of technology. Sleep is fundamental to optimal functioning during childhood, including health and behavior. The purpose of this study was to explore bedtime electronic use and its impact on 3 health consequences-sleep quantity and quality, inattention, and body mass index. Parents of 234 children, ages 8 to 17 years, were surveyed to quantify hours of technology use (computer, video games, cell phone, and television), hours of sleep, and inattentive behaviors. Using any device at bedtime was associated with a statically significant increased use of multiple forms of technology at bedtime and use in the middle of the night, reducing sleep quantity and quality. Little association was found between technology use and inattention. A statistically significant association was found between bedtime technology use and elevated body mass index. Clinicians should discuss the impact of technology at bedtime to prevent harmful effects of overexposure.

The outcomes of simultaneous liver and kidney transplantation using donation after cardiac death organs.

BACKGROUND: There has been a remarkable increase in simultaneous liver and kidney transplantations (SLK). As organ demand has increased, so has the use of donation after cardiac death (DCD). However, little is known about the outcomes of DCD in SLK. METHODS: We performed a retrospective analysis using the United Network for Organ Sharing database to compare the outcomes of DCD SLK to donation after brain death (DBD) and determine the impact of donor and recipient factors on allograft and patient survival. RESULTS: Between 2002 and 2011, a total of 3,026 subjects received SLK from DBD and 98 from DCD. Kidney, liver, and patient survival from DCD donors were inferior to DBD at 1, 3, and 5 years (P=0.0056, P=0.0035, and P=0.0205, respectively). With the use of the Cox model, DCD was a significant risk factor for kidney and liver allograft failure and patient mortality. Recipient factors that were associated with worse allograft and patient outcomes included black race, diabetes, being on a ventilator, hospitalization, delayed graft function, hepatocellular carcinoma, and intensive care unit stay. Older age of the donor was also associated with worse outcomes. CONCLUSION: Despite the decreased allograft and patient survival compared with DBD, DCD SLK provides an acceptable option for SLK, with a survival probability of more than 50% at 5 years.

Responses to a theoretically adapted clinical trial education session: faith-based sites versus rural work site dissemination.

The process for advancing biomedical knowledge depends upon recruiting an adequate and representative sample of individuals to voluntarily participate in research studies. A consistent personal barrier to cancer clinical trial participation is the lack of awareness and understanding related to trial availability, and the prevention and treatment roles participation represents. In particular, comprehensive community-based approaches to recruit and educate rural residents are needed. Moreover, consistent under representation of priority populations should be addressed with innovative outreach to collaborate in identifying culturally meaningful approaches. A theoretically adapted version of a component of the National Cancer Institute's "Clinical Trial Education Series" was assessed via educational sessions delivered through work sites and churches. From eight focus groups with 90 participants, we found that church leaders, congregants, and community members were receptive to education on cancer research, increased their short-term knowledge about it, and intent to participate in cancer studies, decreased their current anxiety about clinical trials participation, and provided specific suggestions for further adapting the educational session to be even more culturally relevant. These outcomes provide evidence to support the effectiveness of future customized recruitment strategies embedded within a community or faith-based environment that may increase knowledge, decrease anxiety and intent to actual participation in cancer studies, as well as impact study representativeness and address causes of health disparities.

Comparison of trauma mortality between two hospitals in Turkey to one trauma center in the US.

OBJECTIVES: The development of comprehensive international trauma case registries could be used to perform outcomes analysis and comparisons between countries with the goal of improving trauma care worldwide. METHODS: A retrospective study (April 2004 to April 2005) of injured patients from a Pennsylvania state trauma center (PSTC) were case matched according to age, sex, and injury severity score with two Turkish hospitals. Patients' demographics (age, sex), prehospital information (mechanism of injury, mode of transportation), injury severity (injury severity score and Glasgow coma score), and outcomes (intensive care unit length of stay, hospital length of stay, mortality) were collected. STATISTICAL ANALYSIS: P value of less than 0.05, odds ratio (OR), chi2 test, two-sample t-test, mean+/-SD. RESULTS: Medical records from 506 Turkish trauma patients were abstracted and compared with 506 injured patients in the PSTC registry. Patients in Turkey presented more commonly with a Glasgow coma score of less than or equal to 8 (13.09 vs. 4.26%, P<0.01, OR 3.38) had increased mortality (8.30 vs. 0.79%, P<0.01, OR 11.36) and required mechanical ventilation more than 1 day more often (16.44 vs. 8.75%, P<0.01, OR 2.05). Motor vehicle crashes were the leading cause of injury in both groups. Assaults and falls were more frequent in the PSTC. Pedestrian injuries were more common and had higher mortality rates in Turkey (P<0.05). CONCLUSION: This study demonstrates significantly worse outcomes in trauma care and higher mortality rates in Turkey versus PSTC. Developing a trauma registry to monitor improvements in patient care and to target injury prevention strategies should be a high priority for the Turkish healthcare system.