Artificial Intelligence (AI) is a broad area of computer science involving technologies that can learn to make decisions and perform tasks that previously required human action.1 At its core, AI depends on two things – data and the models used to train algorithms to behave in certain ways based on the data. AI can manipulate the algorithms by learning behavior patterns from the data set. Greater quantities and higher quality data sets result in higher performance from the AI.
Some AI applications, such as facial recognition software, social media algorithms, and self-driving cars have gained notoriety and raised legal and ethical concerns. More commonly, however, AI is being used to accomplish undertakings impractical for humans, such as discerning patterns and predicting outcomes from massive amounts of data, or to perform routine tasks that take valuable time from workers whose talents might be put to better use.
Some AI applications drive common consumer experiences, such as personalized shopping recommendations, chatbots and voice assistants, spam filters, and GPS technologies. Businesses use AI for employee recruitment, security and fraud prevention, transportation logistics, and personalized marketing, to name just a few examples.
Many AI applications are built on foundational, or large language models, which can be adapted to a wide range of downstream tasks. Generative AI systems, such as ChatGPT and GPT-4, are being used to generate foundational content for reports, articles, scripts, and even musical scores.
AI technologies are advancing quickly, and the use of AI in business has been growing steadily. Since 2017, the number of organizations that report using AI has increased 2.5 times.3 Many, industries are using AI in some form today, as well as planning to expand its applications going forward, including the healthcare and workers’ compensation sectors.
Currently, AI is being used by hospitals and healthcare systems in multiple ways, including adverse-event prediction, schedule optimization, and inventory control.4 Healthcare providers are using AI for such applications as chronic disease management and personalizing the patient experience. However, overall adoption is still in early stages. Only 14% of healthcare professionals report using AI today, and 33% believe that AI will hurt the healthcare industry more than it helps.5 Industry analysts disagree, however. According to McKinsey, “AI represents a meaningful new tool that can help unlock a piece of the unrealized $1 trillion of improvement potential present in the [healthcare] industry.”4
Projections for the future of AI in healthcare are optimistic. The global market for healthcare-specific AI applications was valued at $16.5 billion in 2022 and is expected to reach $198 billion by 2030.6 The anticipated rapid growth is attributed to multiple factors, including the COVID-19 pandemic (which put excessive pressure on the healthcare system and accelerated the need for automated solutions), increasingly large volumes of patient information available through electronic health records (EHR) and other data sources, and expanding opportunities to develop AI applications using the revolutionary ChatGPT and GPT-4 technologies.
There is no question that the healthcare industry needs solutions to combat staffing shortages and rising costs. AI holds the promise of significant help, as well as some risks.
As noted, AI is already being used in healthcare but is nowhere near reaching its full potential. New and expanded ways that AI might improve healthcare delivery abound, including:
Diagnostics and Imaging: AI’s ability to process large amounts of data and identify patterns, as well as recognize complex details within images, such as X-rays and MRIs, has already succeeded and holds even greater promise for, assisting radiologists in making correct diagnoses. To date, the Food and Drug Administration has approved approximately 400 AI algorithms for the radiology field.7
Clinical Decision Support: AI holds great promise for improving clinical decision support at the point of patient care. By analyzing vast amounts of patient data to find patterns, AI algorithms can assist medical professionals with identifying warning signs and making more informed decisions about treatment options. Machine learning based early warning systems, such as Epic’s Deterioration Index (EDI), are already performing better than traditional early warning methods, such as MEWS (modified early warning score), and are expected to continue improving.8
Medication Management: AI can be used to analyze clinical data and identify patient subgroups that are more likely to respond well – or not so well — to a drug and improve the chances of prescribing the optimal therapy and dosage without relying on trial and error. New and better drugs may also become available more quickly and cheaply by using AI in the drug discovery process. Pharmaceutical researchers can draw on data related to the chemical properties of existing drugs to generate new drugs and predict the safety and efficacy of new drug candidates by analyzing clinical trial data. Gartner predicts that by 2025, more than 30% of new drugs will be systematically discovered using generative AI techniques.9
Remote Care: AI is already facilitating remote care in some ways and as the technology advances, more patients may be able to receive quality remote care. Wearables, which are portable medical devices that monitor various health indicators, such as heart rate and blood pressure, will continue to be improved by AI better detecting and correcting errors and outliers, and extracting relevant information from images or sequences. Going forward, AI will assist with suggesting diagnoses by drawing on large stores of relevant data, automatically querying and instructing patients, and creating reports for the medical record. AI can also increase access to care by matching provider skill sets with population needs in given locations and automatically routing patients to alternative providers as needed.
Personalized Care: AI can be used to assimilate information about patients’ medical and family histories, genetics, lifestyle, and other social determinants of health, to customize care and provide optimal treatment based on an individual’s unique profile. AI can also facilitate real-time translation to bridge language barriers between patients and providers and convert technical medical terms into simpler language for easy comprehension. This combination of contextual care and customized communication could go a long way in improving patient satisfaction.
Operational Efficiencies: The administrative burdens of healthcare are many and have become more acute amid staffing shortages in recent years. AI can help to reduce workloads and improve operations in a number of ways. Digital communication channels, such as mobile apps and virtual assistants (chatbots) driven by more accurate data and natural language capabilities will perform better and reduce reliance on staff to manage appointments and communications. Speech recognition software can enable automated transcriptions of clinical notes into organized summaries for medical records. Coding and billing errors can be automatically identified and corrected, and denials predicted using AI models, before claims are submitted.
For all its benefits, the use of AI in healthcare also comes with questions and risks regarding transparency, accountability, bias, and privacy.
Transparency: By definition, AI algorithms operate under flexible rules and learn new patterns of behavior based on data inputs. This means that they change as they work, which makes it difficult to disclose how they work for purposes of informed consent. Not to mention that the precise functionality of most AI solutions are proprietary trade secrets, so even the healthcare organizations who use them are not likely to be fully informed about how they operate.
Accountability: The adaptive ability of AI software also raises questions of accountability/liability and who is responsible if the AI provides incorrect information. For example, if an AI solution provided 100% correct information for two years, building great confidence among the providers who use it, until one day the AI-generated information is wrong and a provider acts on it, causing patient harm, the question of who is responsible is a muddled one.
Bias: Bias is also a thorny issue with AI algorithms, which have been known to take human and social bias embedded in data and scale it throughout ever-growing applications, which can cause inaccurate risk assessments and flawed decision making. Although the bias begins with the data, it can be exacerbated by the AI training models. A study of one AI algorithm that used spending data to predict severity of illness was found to underestimate – by 50% — the number of black patients who needed greater care. By altering the algorithm to remove the bias, the percentage of black patients who would receive additional help went from 17.7% to 46.5%.10
Privacy: Protecting patient privacy is more challenging with AI, largely because the algorithms depend on data and lots of it. This means that healthcare organizations must share data with partners, usually big tech companies. By law, the use of health information for research purposes requires informed consent from patients, as well as the removal of all identifying information, known as de-identification. However, healthcare data is commonly shared with tech partners without patient consent or knowledge. Worse, algorithms exist for the purpose of identifying individuals in healthcare data repositories, a process known as re-identification. One study found that an algorithm could re-identify 85.6% of adults and 69.8% of children who had participated in a physical activity study, in spite of the fact their protected health information had been removed.11
AI technology is advancing quickly and many concerns about potential misuse and unintended consequences have been raised. As of today, few laws or regulations specific to artificial intelligence exist. However, legislators at the both the state and federal levels are actively investigating the issue and numerous AI-related rules and regulations have been proposed or are pending.
Please see State of the States in this issue for more details
Any improvements to healthcare accessibility, quality, and cost is bound to have a positive impact on workers’ compensation medical care. Reliably accurate diagnostics, better informed clinical decision support, improved medication management, expanded remote care, and higher levels of personalization and efficiency would all positively affect workers’ comp on their own. In combination, the impact could be life changing for injured workers and a game changer for workers’ comp payers.
In addition to the promise of better care, AI may help workers’ compensation medical care programs to:
Automate claims processing: Learning from previous claims data, AI can sort claims according to levels of need, automatically processing the routine claims and routing the small percentage that need claims professional attention.
Classify claims: AI can help to accurately classify injuries by analyzing lines of billing codes to ferret out the primary diagnosis for each claim, which can then be categorized and stored in a database for future intelligence. AI can also extract other information, such as codes for comorbidities, to identify health risks and cost drivers.
Identify outliers: By learning from large data sets, AI can identify outliers beyond cost, including unusual courses of treatment, drug regimens, or duration of therapy when compared to similar injuries and patient characteristics.
Select providers: Predictive models can be used to identify top performers based on relevant and statistically significant performance indicators for each claim’s individual characteristics.
Predict risk and enable intervention: AI can detect when a claim is moving out of the expected cost range and send alerts as soon as a deviation from anticipated trajectory occurs to allow early intervention and prevent further escalation.
Improve fraud detection: AI can scrutinize claims at a greater level of detail to identify hidden indicators of fraud, waste, and abuse.
Prevent litigation: AI algorithms can predict the likelihood of litigation by analyzing and detecting patterns in past claims. At-risk claims can immediately be tagged and routed for special processing to avoid legal issues.
This is hardly an exhaustive list. AI technology is advancing rapidly with new applications being discovered every day, not only to the medical aspects of claims, but to all of workers’ compensation claims management. Exactly which applications will be developed and succeed in the near future remains to be seen. Exactly which applications will be developed and succeed in the near future remains to be seen.
