AI deployments accelerating across an array of complex use cases

Coming through the pandemic, healthcare organizations are ramping up their use of digital technology as they redefine healthcare delivery. The rapid adoption of telehealth in crisis demonstrates their ability to go further.

There have been accelerated shifts toward other emerging healthcare models, along with the investments technologies they require. One quickly growing technology in healthcare is artificial intelligence.

Time-pressured decisions have highly consequential outcomes on a minute-by-minute basis. Using rules-based systems machine learning algorithms, automation can facilitate process execution, drawing on specific patient histories to improve treatment.

By applying AI to these functions, organizations can expedite prior authorization, identify fraud waste, automate billing, coding patient scheduling.

Sashi Padarthy is assistant vice president at Cognizant Healthcare Consulting, leads digital strategy transformation services. Healthcare IT News sat down with him to get his expert views on these aforementioned subjects on the accelerated use of AI in healthcare overall.

Q. How has the use of artificial intelligence in healthcare been accelerated in recent years, including by COVID-19?

A. AI as a technology its overall adoption have significantly advanced over the last few years will continue to accelerate. In the last 15 months we have seen computational biology coming to the forefront.

To be fair, computational biology has been rapidly developing over the last decade as the healthcare industry has gotten access to large datasets, more advanced analytical capabilities, modernized data ecosystems, enabling us to conduct faster more efficient drug discovery research to create precision drugs treatments. COVID-19 vaccinations (Pfizer Moderna) are clear examples of that.

As the medical community has seen the capabilities of AI to help drive this research shorten the amount of time it takes to develop new treatments, the trust reliance on AI is growing.

Computational medicine – the use of AI, machine learning other technologies for early detection diagnosis of disease – has been in the works for nearly a decade. The next milestone is to bring computational medicine to the bedside to be able to identify patient-specific treatments drugs provide them to the bedside clinician in almost real time to significantly enhance patient care.

AI as a technology has advanced in three ways: 1) pattern recognition (computer vision), 2) natural language understanding, 3) natural language generation.

With these advances, a variety of healthcare challenges can be addressed. Here are some examples:

More than 150,000 deaths in the U.S. are related to lung cancer, making it one of the leading causes of death. There are now deep learning algorithms that can detect as well as, or sometimes better than, a radiologist can.

Provider burnout the desire to remove some of the administrative burden has led to clinicians embracing AI. Many providers are now using AI to assist them in creating clinical documentation. AI listens to a patient doctor’s conversation creates a clinical document, which the doctor reviews edits before signing off the chart. This saves clinicians a lot of time increases the accuracy of documentation.

Finally, clinicians are being asked to integrate a tremendous ever-growing amount of data from various EHRs patient-generated data into clinical practices. Until recently we really haven’t had the tools to harness that vast quantity of information in a meaningful way to help patients. AI helps solve that problem. We are selectively using AI now to forecast the spread of different flu strains other contagious diseases a week in advance, with more than 90% accuracy.

Even though AI promises many benefits, we still need to ensure there isn’t any algorithmic bias. Bias is not new in the industry or in healthcare, but, because of its ability to scale, AI can amplify the impact of bias. Therefore the application of AI in a clinical setting will require significant clinical trials to create an evidence base physician buy-in.

Q. Time-pressured decisions have serious outcomes on a minute-by-minute basis. What are a couple of examples of these decisions in healthcare where AI can help?

A. AI has been proven for many use cases is able to assist clinicians in making critical patient-care decisions. AI is not replacing the clinician, nor is it making the decision for the clinician. AI is generating insights for clinicians from data sources traditionally unavailable to a provider at the point of care.

One such example is the use of an AI algorithm to predict psychiatric diagnoses using data from Facebook. The AI algorithm was able to predict psychiatric diagnosis comparable to that of a standard clinical PHQ-9 survey given to a patient to assist the clinician in diagnosing, quantifying or monitoring symptoms severity of symptoms of depression.

The significance being that the questionnaire may have high false-positive rates in primary care settings. Specifically, one meta-analysis found that only 50% of patients screening positive had major depression (Levis 2019). The algorithm, however, has access to large volumes of data that may span days, months years, is objective in its analysis.

Another example: Vocal biomarkers for prediction of psychiatric diagnosis – another AI-driven diagnostic tool for clinicians that can be used by the patient to track analyze over longer periods of time to aid in diagnosing or assessing the severity or change in symptoms of depression.

Access to data outside the EHR, coupled with EHR claims data, are more traditionally available. They are shining a light on use cases that allow the prediction of disease risk, as opposed to diagnosis of an active disease process.

By leveraging AI to analyze larger datasets that include both clinical social data, clinicians can predict a patient’s risk of developing specific conditions, disease processes or suffering a major medical event. It also allows clinicians to develop a patient-specific treatment plan based on the specific health disparities that patients face.

AI algorithms can provide insights to the clinician alerting them that a patient has an elevated risk of developing cardiovascular disease. The algorithms can also provide insight into the challenges the patient faces in mitigating their risk, such as the walkability of their physical environment, access to healthy foods or the quality of care within the geographical area available to the patient.

With added insights into the challenges a patient faces, the provider can work proactively with the patient to determine the most appropriate treatment plan for that individual patient in order to mitigate the patient’s risk factors. Limited to just the information available within the EHR, providers are not able to garner the same level of insights that allow them to provide whole-person care.

Another use-case for AI enabling providers to make faster better-informed decisions is supporting the decision-making process for medically or surgically complex patients. By using deep learning AI machine learning, a provider could weigh the risks benefits of treatment options.

Patients with complex care needs typically have a long medical history, multiple diagnoses multiple comorbidities, which make synthesizing all the information determining a treatment plan based on the best possible outcome difficult time-consuming. A provider could save time determine a better statistical analysis of the risk or benefit of a treatment option that is based on the unique history, diagnosis comorbidities of an individual patient.

AI brings more information to a provider in real time to assist with making difficult complex medical decisions. Providers can leverage key insights at the point of care from multiple data points that are not traditionally available to help patients achieve better outcomes.

Q. Using rules-based systems machine learning algorithms, automation can enable process execution, drawing on patient histories to improve care. Would you elaborate?

A. AI provides the ability to scan across spaces places of care to identify the information that is most relevant to a provider at any given time. Many patients see multiple providers prior to getting to the correct specialist for a specific medical problem. This means that their care the documentation of that care may exist across various clinics, hospitals or health systems.

It is challenging time-intensive for a specialist to have to review various encounter notes, diagnostic testing results other documentation to help them care for the patient. AI can remove that burden learn to identify the specific types of information that a particular provider searches for uses help develop a perspective about the patient based on information from multiple sources.

An example: Cognizant’s Cognitive Computing Data Sciences Lab tackled the challenge of diagnosing diabetic retinopathy (DR) for patients in India who did not have coverage or access to quality eye care. Cognizant partnered with a Bangalore-based clinic, Vittala International Institute of Ophthalmology, to help patients who did not have access to quality equipment specialists.

Cognizant VIIO developed deep learning algorithms that could identify DR 90% of the time, even in low- to poor-quality images. Clinicians upload images the software uses the deep learning algorithm to identify DR. This provides increased access because patients no longer must travel to see a specialist or pay the additional cost of seeing a specialist.

With new models of care coming to the market the tendency for healthcare consumers to shop around, patients will be receiving care across multiple spaces places of care. AI can help create a more unified, seamless experience for the provider patient.

Q. By applying AI to healthcare clinical business functions, provider organizations can expedite prior authorization, identify fraud waste, automate billing, coding patient scheduling. Please explain how this can be done with AI.

A. AI can remove some of the administrative burden surrounding prior authorization, billing coding. AI is better at identifying patterns than a human being. Where a rules-based engine requires updating changes to maintain accuracy, AI learns can get smarter more efficient at recognizing patterns for billing fraud.

AI can not only identify the patterns of fraud faster, but can also help to prevent it. It is capable of sorting through much larger amounts of data identifying patterns of upcoding, whether appropriate documentation exists for a service a patient was billed for other things that seem out of place. 

Like using AI in clinical care clinical decision-making, it is not a replacement for a comprehensive fraud detection program, but a tool to alert a team sooner to patterns that seem out of place.

Natural language processing natural language understanding are being used to assist clinicians with generating clinical documentation by listening to their interaction with a patient turning what it heard into a clinical note. Using this type of AI, a more complete more comprehensive document in a narrative style is created for the provider.

These clinical documents created using NLP NLU are able to show the thought process of the provider more clearly better demonstrate the medical decision-making process a provider went through. 

With CMS finalizing the 2021 Physician Fee Schedule with updated E/M codes, this will better enable providers to code bill for the additional time spent with more involved medical decision-making for complex patients.

AI can remove some of the administrative burden of documenting the necessary information to show the complexity of the patient’s problems medical decision-making for the new billing coding requirements. This also leads to better coding, fewer denials fewer rejected claims.

Like using AI across spaces places of care to help providers gather the most pertinent information, it can also be used to harvest the appropriate information for prior authorizations allow for automatic approvals. This removes the burden on the provider or their staff to have to manually fill out or input the information for prior authorization.

AI-enabled scheduling software can make both a patient’s the medical team’s lives much easier. AI is able to determine the scheduling preferences of a patient by analyzing their past scheduling patterns either auto-schedule or suggest the most appropriate date time, location, provider. 

It can also help to create an optimized schedule for a patient who is seeing multiple providers or receiving multiple treatments in the same day. This reduces the amount of time spent by both the patient the scheduling staff to get the next appointment or series of appointments created.

AI is also able to recognize the complexity of a patient can be used to determine the appropriate visit length for the next appointment. A standard wellness visit for a patient with multiple diagnosis takes longer than one for a patient who has continually been healthy. 

A provider’s schedule can be optimized with this approach because the time allocated to each patient is more customized based on their needs [it] allows the provider to spend the necessary time with each patient without feeling rushed or falling behind in their clinic schedule.

AI is able to make processes that are task-heavy or time-consuming for humans much easier by reducing the overload.

Twitter: @SiwickiHealthIT
Email the writer: [email protected]
Healthcare IT News is a HIMSS Media publication.

Source link

Research suggests Epic Sepsis Model is lacking in predictive power

A new study in JAMA Internal Medicine found that a sepsis prediction model included as part of Epic’s electronic health record may poorly predict sepsis.

Using retrospective data, University of Michigan Medical School researchers found that the predictor did not identify two-thirds of sepsis patients.  

“In this external validation study, we found the ESM to have poor discrimination calibration in predicting the onset of sepsis at the hospitalization level,” UM researchers wrote.   

Epic disputed the study’s findings, saying that the authors used a hypothetical approach that did not take into account the analysis required tuning that needs to occur prior to real-world deployment to get optimal results.  

“In their hypothetical configuration, the authors picked a low threshold value that would be appropriate for a rapid response team that wants to cast a wide net to assess more patients,” said a statement provided by the company.  

“A higher threshold value, reducing false positives, would be appropriate for attending physicians nurses,” it continued.  


As the researchers note, early detection treatment of sepsis have been associated with less mortality in hospitalized patients.

One of the most widely implemented early warning systems for sepsis in U.S. hospitals is the ESM, a penalized logistic regression model included in Epic’s EHR.   

Although Epic developed validated the model based on data from 405,000 patient encounters, the researchers raised concerns about its opacity as a proprietary model.  

“An improved understanding of how well the ESM performs has the potential to inform care for the several hundred thouspatients hospitalized for sepsis in the U.S. each year,” wrote the researchers.

Using the data of all patients older than 18 admitted to Michigan Medicine between December 6, 2018, October 20, 2019, researchers found that sepsis occurred in 7% of the hospitalizations. The ESM had a hospitalization-level operating characteristic curve, or AUC, of 0.63 – “substantially worse,” than that reported by Epic, they said.

When alerting at a score threshold of 6 or higher, which is within Epic’s recommended range, the model identified only 7% of patients with sepsis who were missed by a clinician.  

It did not identify two-thirds of patients with sepsis – despite generating alerts on 18% of all hospitalized patients, creating a large burden of alert fatigue.  

In its statement, Epic argued that the purpose of the model is to identify harder-to-recognize patients who otherwise might have been missed. It pointed to previous research that found the model could accurately predict sepsis, said customers have “complete transparency” into the model.  

According to Epic: “Each health system needs to set thresholds to balance false negatives against false positives for each type of user. When set to reduce false positives, it may miss some patients who will become septic. If set to reduce false negatives, it will catch more septic patients, however it will require extra work from the health system, because it will also catch some patients who are deteriorating, but not becoming septic.  

“In the example given in this paper, if the Epic model was used in real time, it would likely have identified 183 patients who otherwise might have been missed,” the statement added.  


Health systems have increasingly turned to machine learning predictive analytics to detect sepsis in patients in an effort to decrease mortality.  

In 2019, researchers from Geisinger IBM developed a new predictive algorithm to detect sepsis risk, aimed at helping clinicians create a more personal care plan for at-risk patients.  

But the JAMA study reiterates that models have their own challenges, such as alert fatigue or, conversely, defaulting to computer-generated assessments as infallible.  


“Medical professional organizations constructing national guidelines should be cognizant of the broad use of these algorithms make formal recommendations about their use,” wrote researchers.


Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: [email protected]
Healthcare IT News is a HIMSS Media publication.

Source link

1 2,396 2,397 2,398