“AI in Health Care: Data types and AI models” is my notes from the course: “Startup Opportunities using AI in Healthcare (Updated Dec 2021),” An Online Skillshare Class by Dr. Saurabh Bhatia. There are a lot of benefits of the application of AI in health care.
Benefits for patients:
- Better care
- Cost effective co-pay models: AI can help patient in deciding a better co-pay models based on risk factors of the patient
- lesser complications/recurrences
Benefits for care providers and stakeholders:
- Lesser re-admissions/workload
- Higher performance of providers
- Cost-saving interventions increase the profits
Clinical outcomes depend not only on the treatment given by healthcare personnel inside the hospital but also on the conditions outside the hospital or a healthcare facility. For example, clinical outcomes or the risk of contracting a disease also depend on the living conditions, socio-economic status, nutritional status, etc. We can train AI algorithms based upon this data to suggest specific interventions based on these factors. This concept gives rise to two different solutions: In-Hospital and out of the hospital solutions.
Dr. Saurabh takes about a scenario describing two patients with an identical health problem – one with good family support and sufficient money to afford treatment and another person with no family support and trying hard to make two ends meet. The various possible outcomes for either of them are recovery, readmission, complications, disability, or death.
The second person with financial hardships and no social support is likely to get complications leading to disability or death. He is also more likely to get readmitted due to complications or recurrences. This further increases the burden on the patient. This may also decrease patient satisfaction at that particular hospital.
Insurance companies tend to increase the premia for certain diseases, looking at the poorer patient outcomes of a particular disease. There might be a decrease in the reimbursement rate for providers. This adds more burden to the patient as well as health care providers.
AI in health care applications
Artificial intelligence or machine learning models can be used in different scenarios. The use cases of these AI models can be broadly divided into Inside hospital applications and out of the hospital applications.
Inside the hospital
- CDSS (clinical descision support systems): New information can be collated using neural networks and brought into clinical practice.
- Rules engines: When certain values or information is entered into electronic medical record, action is taken based upon pre-existing rules. For example, advising a patient to consult at an obesity clinic if BMI is greater than certain number. These rule engines can be updated by incorprating with AI algoritms so that better recommendations can be made based upon the various factors influencing the health of a patient.
- Automated Documentation: Artificial Intelligence can created automated documentation such as discharge summary etc based upon the case record of the patient.
- Claim rejection prediction: Many Insurance claims get rejected duue to incomplete documentation or poor quality of documentation. AI can be used to help healthcare providers produce complete documentation so as to minimise the chances of rejection of an insurance claim.
- Hospital stay prediction: AI algorithms can help predict the number of days of hospitalisation based on the current condition. This will help the hospitals plan their resources so as to provide the best possible care to the patients.
- Cost of the treatment prediction: Predict the approximate cost of the treatment including the additional expenses inside and out of hospital.
- Device use prediction: AI can be used to predict the number of days a patient may use a particulare device in the hospital
Out of hospital
- Ambulance: AI can be used to determine the specialists to be sent along with an ambulance.
- Pharmacy: AI can be used to increase effeciency of pharmacy stock management, pharmacy costs, demand of a particular medicine.
- Food: AI algoritms can predict the outcomes of a particular condition depending on the food or nutritional status of a patient.
- Complication prediction: AI algorithms can be sued to predict the complications that might occur and that chance of those complications in a particular patient.
AI in health care data requirements
- Clincial data: Can be obtained from electronic medical records
- Logistic data: easy of acccess of medical facility, availability of care givers
- Financial data: Affordability of a service at a health facility. Is the patient insured or self-paying? if he is insured which plan he is on?
- Family data: Non-modifiable genetic data, living conditions, attitude and psychology of the patient.
If an AI/ML system is trained with adequate high-quality data, it can help in the following:
- Identification of high risk patients
- early identification of intervention points
- Intervention recommendations
- Collection and analysis of results when intervention was done vs when not done. This data can be used to show the effcetiveness of an AI system as well serves as a feedback to the AI/ML system
- Improved patient outcome
- Descriptive models: Learning from the past
- Answer the question “what has happended?”
- Use data mining and aggregation to provide insights
- AI can be used to find deficiencies in health records by enabling AI to comapre a health record with other health records.
- Predictive models: Predicting the future possibilty
- Answer the question “what could happen?”
- use statistical models and forecast techniques to understand the future.
- eg: using prognostic data of a particular disease
- Prescriptive models: Recommending an intervention
- Answer the question “what should we do?”
- use optimisation and simulation algorithms to advise on possible interventions and their outcomes.
- AI can be used to suggest a particular intervention at a right time so as to prevent the occurence of a high risk condition.
Challenges to implementation of AI in health care
- Disengaged patients: People may not reveal their true data – for example, financial conditions, co-morbidities. Patients might not be interested to reveal an accurate financial condition. Patients might not reveal their co-morbidities unless asked.
- Incomplete data: Improvemnt of EMRs/ software systems to include social and financial data to make predictive and prescriptive data models. But collecting this data increases the amount of time an dhospital resources spent per person.
- Technology upgarde barrier: cloud providers can be used to decrease the cost of upgrading to new software or hardware. Health personnel should be trained to incorporate AI into their clinical practice.
Check out other posts on sciencetidings.com Subscribe to the med-dose newsletter