Practicing physicians these days are barraged with a lot of technical jargon promoted by the Information technology professionals such as Big Data, Hadoop, Artificial Intelligence and Predictive analytics. For a physician not introduced to the these terms, the conversation is of little value unless there is a specified value in the clinical setting.
Pulitzer Prize winning author Siddhartha Mukherjee in his book ‘Emperor of Maladies’ has quoted “The greatest clinicians who I know seem to have a sixth sense for biases. They understand, almost instinctively, when prior bits of scattered knowledge apply to their patients—but, more important, when they don’t apply to their patients.”
So the accumulation of all the 6th sense of multiple physicians in the form of scattered notes in documents is the food for Big Data professionals, who curate this data manually with standard vocabularies and then analyse patterns that can help physicians make informed decisions at the point of care.
While Predictive Analytics is a huge winner if there is genomic data available, the fact remains that in current clinical scenarios, complete genomic sequencing is not yet the norm. They are still restricted to the areas of research as the cost of doing a complete sequencing of the genome is still about $ 1000 and very few institutions are prescribing it as a norm in routine clinical practice. Once that is in place, it can lead to a whole lot of practical applications and use cases making the use of technology like Big Data and Hadoop.
Till then, use cases which use traditional computing powers like SQL and simple queries and algorithms running on top of it could be used to get the benefits of clean data and technology in the clinical workplace.
In an attempt to utilize Predictive Analytics and Artificial Intelligence in the world of EMR and healthcare data, a list of 5 practical clinical use cases and 5 administrative/Claims Related use cases of how data could be used at the point of care and integrated with the EMR is listed out below:
Pulitzer Prize winning author Siddhartha Mukherjee in his book ‘Emperor of Maladies’ has quoted “The greatest clinicians who I know seem to have a sixth sense for biases. They understand, almost instinctively, when prior bits of scattered knowledge apply to their patients—but, more important, when they don’t apply to their patients.”
So the accumulation of all the 6th sense of multiple physicians in the form of scattered notes in documents is the food for Big Data professionals, who curate this data manually with standard vocabularies and then analyse patterns that can help physicians make informed decisions at the point of care.
While Predictive Analytics is a huge winner if there is genomic data available, the fact remains that in current clinical scenarios, complete genomic sequencing is not yet the norm. They are still restricted to the areas of research as the cost of doing a complete sequencing of the genome is still about $ 1000 and very few institutions are prescribing it as a norm in routine clinical practice. Once that is in place, it can lead to a whole lot of practical applications and use cases making the use of technology like Big Data and Hadoop.
Till then, use cases which use traditional computing powers like SQL and simple queries and algorithms running on top of it could be used to get the benefits of clean data and technology in the clinical workplace.
In an attempt to utilize Predictive Analytics and Artificial Intelligence in the world of EMR and healthcare data, a list of 5 practical clinical use cases and 5 administrative/Claims Related use cases of how data could be used at the point of care and integrated with the EMR is listed out below:
Clinical Usecases
1 Acute Kidney Injury
According to Statistics, 1.2 million people per year get AKI during a hospital stay and 300,000 people in the US die annually due to AKI, this is more than breast cancer, prostate cancer, heart failure and diabetes combined.
The successful documentation and implementation of the AKI algorithm published by the NHS is a systematic and step by step program that can help reduce the incidence of Acute Kidney Injury and the mortality and comorbidities associated with it. The e-alert for AKI installed in the LIMS (Laboratory Information Management Systems) alerts physicians when a patients’ Serum Creatinine rises greater than 26 mmol/litre from a baseline within 48 hours or there is a rise of 50% or more in 7days and/or the urine output is < 0.5 ml/kg body weight/hour. https://www.england.nhs.uk/wp-content/uploads/2014/06/psa-aki-alg.pdf
The successful documentation and implementation of the AKI algorithm published by the NHS is a systematic and step by step program that can help reduce the incidence of Acute Kidney Injury and the mortality and comorbidities associated with it. The e-alert for AKI installed in the LIMS (Laboratory Information Management Systems) alerts physicians when a patients’ Serum Creatinine rises greater than 26 mmol/litre from a baseline within 48 hours or there is a rise of 50% or more in 7days and/or the urine output is < 0.5 ml/kg body weight/hour. https://www.england.nhs.uk/wp-content/uploads/2014/06/psa-aki-alg.pdf
2 Sepsis Management
Sepsis or SIRS (Systemic Inflammatory Response Syndrome) accounts for 20 to 30% hospital deaths and $15.4 billion in annual healthcare costs. Needless to say early diagnosis and treatment is critical to optimal care.
In a study, it was found that 24% of infected patients with 2 or 3 qSOFA (or Quick Sepsis Related Organ Dysfunction Assessment) points accounted for 70% of the deaths. An automated EMR based Sepsis identification system helps to detect cases with sepsis. A sepsis sniffer algorithm in ED identifies patients who exhibit at least 2 of the 3 criteria of Altered Mental Status, Fast Respiratory Rate and Low Blood Pressure and initiates the Sepsis Order set if the results of the score warrant it. The content on the Sepsis Order Set helps to make sure that none of the key variables for the management of sepsis is missed out.
https://www.mdanderson.org/documents/for-physicians/algorithms/clinical-management/clin-management-sepsis-management-adult-web-algorithm.pdf
In a study, it was found that 24% of infected patients with 2 or 3 qSOFA (or Quick Sepsis Related Organ Dysfunction Assessment) points accounted for 70% of the deaths. An automated EMR based Sepsis identification system helps to detect cases with sepsis. A sepsis sniffer algorithm in ED identifies patients who exhibit at least 2 of the 3 criteria of Altered Mental Status, Fast Respiratory Rate and Low Blood Pressure and initiates the Sepsis Order set if the results of the score warrant it. The content on the Sepsis Order Set helps to make sure that none of the key variables for the management of sepsis is missed out.
https://www.mdanderson.org/documents/for-physicians/algorithms/clinical-management/clin-management-sepsis-management-adult-web-algorithm.pdf
3 Initiation of Statins
Mayo clinic has implemented a single click decision support tool within the EMR to automate the calculation of 10-year atherosclerotic cardiovascular risk and populate the statin choice decision. Based on the risk score calculation parameters of the ASCVD (atherosclerotic cardiovascular disease), the tool populates whether the patient needs a statin or not and if yes, which statin would be the best choice. This is defined on the basis of an algorithm running which has historical data of similar patients along with their outcomes.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6157431/
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6157431/
4 Algorithm to identify major cardiac adverse events while on statins
Cardiovascular disease is a leading cause of death worldwide and statins are largely prescribed for the same. A study was conducted using historical data to identify patients with MACE (Major Adverse Cardiac Events) while on statins for primary prevention. This algorithm achieved a 90 to 97% positive predictive value for the identification of adverse cardiac events while on statins
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4333709/
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4333709/
5 Detection of Diabetic Retinopathy
Diabetic retinopathy is the fastest growing cause of blindness and screening patients with diabetes for retinopathy has been mandated as part of all the Quality Measures such as MIPS, HEDIS and others. Google analysed a dataset of 128,000 images of the retinal fundus and used that to predict retinopathy based on the lesions present on the scan such as microaneurysms, haemorrhages, exudates etc. A blindfold comparison of the results predicted by the AI algorithm was compared with the results given by experienced ophthalmologists. The AI algorithm had an accuracy of 90% which is a huge win especially in areas where there is a shortfall of specialists to interpret the results.
In the areas of diagnostics especially interpretation of Imaging results, a lot of good work is going on and soon AI capabilities will lead decision making in the interpretation of reports.
https://ai.googleblog.com/2016/11/deep-learning-for-detection-of-diabetic.html
In the areas of diagnostics especially interpretation of Imaging results, a lot of good work is going on and soon AI capabilities will lead decision making in the interpretation of reports.
https://ai.googleblog.com/2016/11/deep-learning-for-detection-of-diabetic.html
Physicians working closely with Data scientists can help create some exciting algorithms given that healthcare centres today have large volumes of data sitting in their EMRs. A physician with even 50 patient records of 2 years has sufficient information to derive useful insights on predicting interesting patterns on the future health and preventive measures therefore can be planned accordingly. The approach for healthcare institutions should be to first identify their most pressing problems and then evaluate if any kind of Prediction of solutions to those problems could lead to better outcomes in patient care. A well thought and organized approach could go a long way to develop algorithms that work well for the hospitals and ambulatory clinics to achieve their goals and have favourable outcomes for both the patient and the hospital.
Administrative Use Cases:
In Patient Registration systems and billing systems, the algorithms can have a different flavour, some of which are listed below:
1. Healthcare Utilization Management: Integration of hospital resource utilizations in different departments, risk management and quality assurance into a management dashboard in order to ensure the judicious use of the facility's resources. A review of the procedures and services rendered by the hospital and the resources used by them (such as rooms, timings, instruments, admin and clinical staff) can help to identify under utilized and spare capacity of the hospital. This can be inbuilt in the system with the help of algorithms so as to get an optimal effect.
2. Capturing Missed and Incorrect Charges: Many facilities lose revenue due to billing errors done manually. Billing amounts associated with each DRG (Diagnosis Related Group) can be used to create models and compare them with the actual billing. Outliers in patient invoices can help to identify the likelihood of missing or incorrect charges. The higher prioritized invoices can then be reviewed to confirm the charges, which may be incorrect, over billed or under billed.
3. Predicting denials: There is a constant tug of war between the providers and payers about the claims to be settled. Most of the denials are due to incorrect documentation or missing information, duplicate claims, service already paid as a part of bundled services or others, services not covered by payer and late submission of claims. All of the above can easily be avoided by streamlining coding and billing processes. The industry benchmark for medical billing denials is 2% and in practices between 5% to 10%. Reworking on denials has a cost associated, so having algorithms running within the systems can help to streamline these efforts.
4. Predicting wait times and No shows: Idle time in clinics, both for patients and providers can be frustrating. In departments where procedures are done, it is possible to predict treatment durations by looking at historical data. Similarly the number of no shows of scheduled appointments can be predicted by including variables like age of patient, severity of disease, previous appointment regularity of the patient and others. This could help the facility to take in additional patients on the free slots proactively.
5. Measuring Bad debt, Days in A/R and DNFB (Days in Total Discharged Not Final Billed): These metrics help to determine the effectiveness of the claims generation process and effectiveness of collection efforts.
A number of KPIs tracked by hospitals and payer systems can be in built within the Revenue Management Systems and Scheduling systems. These include patient costs, ROI of the facility, Average Patient Wait times, Average patient appointment scheduling times and Patient Satisfaction.
With the healthcare market growing and many EMRs, both locally created and branded companies in the play, it is eventually the content within the EMR that can make it user friendly. Many content companies offer excellent standardised algorithms and content that can be integrated within the EMR, but there always remains physician preferences based on specialty, region of practice and other personal preferences. The end user i.e. the physician ultimately needs to decide on what suits his practice and then have the algorithms accordingly tailored within the EMR.
1. Healthcare Utilization Management: Integration of hospital resource utilizations in different departments, risk management and quality assurance into a management dashboard in order to ensure the judicious use of the facility's resources. A review of the procedures and services rendered by the hospital and the resources used by them (such as rooms, timings, instruments, admin and clinical staff) can help to identify under utilized and spare capacity of the hospital. This can be inbuilt in the system with the help of algorithms so as to get an optimal effect.
2. Capturing Missed and Incorrect Charges: Many facilities lose revenue due to billing errors done manually. Billing amounts associated with each DRG (Diagnosis Related Group) can be used to create models and compare them with the actual billing. Outliers in patient invoices can help to identify the likelihood of missing or incorrect charges. The higher prioritized invoices can then be reviewed to confirm the charges, which may be incorrect, over billed or under billed.
3. Predicting denials: There is a constant tug of war between the providers and payers about the claims to be settled. Most of the denials are due to incorrect documentation or missing information, duplicate claims, service already paid as a part of bundled services or others, services not covered by payer and late submission of claims. All of the above can easily be avoided by streamlining coding and billing processes. The industry benchmark for medical billing denials is 2% and in practices between 5% to 10%. Reworking on denials has a cost associated, so having algorithms running within the systems can help to streamline these efforts.
4. Predicting wait times and No shows: Idle time in clinics, both for patients and providers can be frustrating. In departments where procedures are done, it is possible to predict treatment durations by looking at historical data. Similarly the number of no shows of scheduled appointments can be predicted by including variables like age of patient, severity of disease, previous appointment regularity of the patient and others. This could help the facility to take in additional patients on the free slots proactively.
5. Measuring Bad debt, Days in A/R and DNFB (Days in Total Discharged Not Final Billed): These metrics help to determine the effectiveness of the claims generation process and effectiveness of collection efforts.
A number of KPIs tracked by hospitals and payer systems can be in built within the Revenue Management Systems and Scheduling systems. These include patient costs, ROI of the facility, Average Patient Wait times, Average patient appointment scheduling times and Patient Satisfaction.
With the healthcare market growing and many EMRs, both locally created and branded companies in the play, it is eventually the content within the EMR that can make it user friendly. Many content companies offer excellent standardised algorithms and content that can be integrated within the EMR, but there always remains physician preferences based on specialty, region of practice and other personal preferences. The end user i.e. the physician ultimately needs to decide on what suits his practice and then have the algorithms accordingly tailored within the EMR.
Author
Dr. Joyoti Goswami
Healthcare expertise of over 20 + years. Clinician having worked in specialty hospitals as medical officer and currently in the Healthcare Information Technology domain since the last 10 + years. Have a good combination of clinical and technology skills in multiple areas. Worked with multiple EMRs such as Allscripts, GE Centricity, Athena health and Nextgen
