Author: Maxwell Jen
Through the Meaningful Use (MU) incentives and universal adoption of electronic health record (EHR) systems, the HITECH (Health Information Technology for Economic and Clinical Health) Act was supposed to improve the quality, and delivery of patient care through clinical decision support, automated safety systems and reduced costs. However, several studies since MU introduction have shown that while EHRs may be associated with increased guideline compliance, improved research capabilities, and increased transparency, EHRs are not consistently associated with better care or outcomes. Instead, EHRs have been associated with worsened patient rapport, highly variable documentation quality, and increased prescribing and order entry times. Several reasons have been offered for these undesired effects. When analyzing patient rapport, investigators noted that physicians spent a significant proportion of the patient evaluation staring at the screen for documentation and computerized physician order entry (CPOE). On documentation time and quality, investigators noted that EHR prompted not only an increased quantity of data entry, thus prolonging documentation time, but also increased redundant and erroneous data entry, resulting in the propagation of EHR-related medical errors. Institutions have tried to counter these errors with pop-up warnings but poor implementation has created what has been termed alert fatigue, resulting in providers, overriding even relevant warnings. Lastly, the fundamental design of a number of EHRs is associated with missed lab results. A well-designed, integrated artificial intelligence (AI) platform combined with machine learning and natural language processing (NLP) applications has the potential to mitigate or entirely avoid some of these issues. In direct patient care, AI has the capacity to generate the entire patient encounter document. Ideally, the provider would bring a mobile device or computer with microphone capability to the bedside. As the physician speaks with the patient, the NLP would transcribe the conversation word for word while the AI extracts relevant data from the transcription and populates it into the encounter chart. Quantitative data could be pulled directly from the clinical laboratory and cardiac monitoring devices. Simultaneously, the AI would cull the available data and generate a report of prior data relevant to the current encounter. For example, if the patient mentions “chest pain” as a complaint, the AI would generate a report composed of cardiovascular risk factors and the most recent ECG and stress test results. Machine learning regression algorithms, with this data, could even generate probability profiles for emergent causes of chest pain e.g. myocardial infarction, pulmonary embolism at the bedside in real-time. Regarding clinical alerts and lab results, AI could similarly cull the data and, based on past data and regression algorithms, assess the potential risk for adverse events and generate an appropriate alert or even no alert if risk is deemed sufficiently low. The passing of HITECH was a seminal event for EHR adoption. However, HER adoption has not yet achieved the goals espoused by its proponents. AI may be the key to unlocking the potential of EHR and fulfill the promise of data-driven, technologically-augmented, value-based healthcare.