State of the Art Technology to tackle the Epidemic of Opioid Crisis
Substance use disorder is undoubtedly an epidemic in the U.S. Since 1999, the number of opioid overdose deaths have quadrupled and reach the figure of 100 deaths per day in 2016. In 2016, according to the US Department of Health and Human Services, around 115 million people misused their prescription pain medications resulting in about 42000 deaths. In the past three years, the figure has risen alarmingly and stands at 70000 deaths in 2019. Over the past ten years, the deaths due to prescription drug abuse have tripled. Studies state that working people who are addicted to prescription drugs for 3 months or more almost never return to work. Around 4.5 million US citizens are said to be addicted to prescription opioids and the US accounts for about 80% of the total opioid consumption in the world.
Treatments for opioid addiction aren’t cheap either. The US government spends about $500 billion annually to provide treatments and therapy for substance abuse patients. From these statistics, it is easy to observe the domino effect caused by opioid addiction and how it impacts everything from personal health to the nation’s economy. Deaths due to drug overdoses now consume more lives in the US than accidents do in India. Legislative acts and passing laws to criminalize addiction only tend to exasperate the problem further. The only way out is to identify drug-seeking behavior in the early stages, proactively prevent and make the human intervention more effective in terms of patient outcomes and cost.
Overview of the role of technology in the Opioid Crisis:
Big Data Analytics and AI have proven their mettle in the early detection of drug-seeking behavior and providing crucial data to healthcare providers who can then identify at-risk patients who have a history of drug dependency before they prescribe to them.
Data Analytics and Machine Learning algorithms have a crucial role to play in confronting the opioid crisis by providing innovative pain management solutions, monitoring strict control over prescription drug abuse and drug theft incidents. AI, Deep Learning Predictive Models, and Predictive Analytics can provide valuable insights into the drug-seeking mindset by analyzing pattern changes in behavior. Continuous monitoring and effective prediction about patients who are more likely to relapse is another niche area where the involvement of technology has yielded improved results by helping providers with data insights for targeted prevention strategies.
Prevention of Opioid misuse and addiction using Technology:
Pattern change detection leads to prevention is the basic principle on which technological helps providers and the government to combat the fight against addiction. By studying three important information it is easy to determine common drug-seeking characteristics — EHR patient data history that has diseases and conditions recorded, the kind of drugs that have been prescribed to the patient, and the behaviors and symptoms exhibited by the patient because of each of the drugs. Not just symptoms but a chain of actions, descriptive questions, or phrases used by the patient along with the related circumstances add up to providing wholesome data that can act as a valuable indicator for current drug addiction or predict the risk of future abuse.
Apart from pattern detection, people seeking drugs to satiate their addiction tend to display certain common characteristics which are used as key indicators to alert healthcare providers. People with a history of clinical depression, anxiety, or PTSD are more likely to become abused medications.
These people might ask for additional prescriptions from their physician by stating that have run out of a particular medication or that a lower dosage doesn’t work for them. Others might describe allergies to certain medications and influence their physicians to prescribe the same medication during each follow-up visit.
Another inherent problem in prescriptions for pain medications is that there is no way to measure pain and that it is extremely subjective. Each patient’s tolerance threshold to pain is different. There are no biomarkers to measure the level of pain experienced other than what is self-reported by the patient. In patients who experience chronic pain, the body naturally develops a tolerance to pain meds after a certain period of time and the same dosage that was prescribed initially might not work later down the line and they might require higher dosages. Considering the addictive nature of opioids, this is a slippery slope, and physicians are faced with the challenge of balancing the pain management needs of the patient and the risk of them getting addicted.
By leveraging patient data in EHR such as reported pain levels and related statistics along with Machine Learning, physicians can develop better pain management solutions. Machine learning can be employed to analyze the feedback from patients based on which a Reinforcement Learning model can be developed which will assist health care providers to develop personalized pain management solutions that adapt to the patient’s changing needs.
Data collected from patient-physician conversations regarding the intensity of pain experienced coupled with the information of population health datasets will make this model automatically personalize the pain management needs specific to the patient. These technological interventions will make pain management less subjective. Once that is done, the detection of anomalies would become a lot easier.
It has been observed that in some states in the US, the prescription written by physicians for pain medications is 3x when compared to others. Machine learning can also detect abnormalities with the frequency of prescriptions for pain meds using EHR data. It can also detect and alert care providers if the patient obtains prescription drugs from two different physicians or uses the prescription to purchase the next course of medications from two or more locations before an acceptable period. When such irregularities are seen, alerts are sent out to care providers and early intervention is enabled.
Physician’s notes — A Gold Mine:
There is nothing more valuable than the physician’s notes which contain the information collected during a direct observation of the patient. Information about the patient’s behavior, appearance, body language, social circle, demographics, general mental well-being, regularity to work along with the physician’s intuition and diagnosis are at the core of AI/ML solutions to identify drug-seeking behavior and enabling early intervention. A Machine Learning solution that collates data from free-flowing physician notes using Natural Language Processing, PHI from EHR, and the patient’s behavioral patterns would be mighty effective in early detection. Phase-based extraction, rule-filtering, and text clustering are certain mechanisms based on which the ML algorithms identify abnormal variations in inpatient data and alert the respective care providers.
Prevention of Addiction Rehabilitation Relapse using Technology:
Opioid addiction treatments are expensive. Each time a patient in rehab relapses, the cost involved compounds. While support group programs such as Narcotics Anonymous are effective, studies show that 75% of the attendees relapse within the first year. AI can be used to identify changes in behavior by monitoring the patient’s routine, sleep patterns, and keyword usage.
These have been identified as key indicators that suggest if a patient has an increased likelihood of relapse. When the AI encounters data that tends to align with the indications of an impending relapse, it sends an alert to the care providers and sponsors. Human intervention is an irreplaceable part of the scheme of things but AI helps the intervention to happen earlier so that it is effective.
Leveraging the data from IoHT Devices:
By integrating Machine learning algorithms with streams of real time data from wearable health monitoring devices, abnormalities can be detected in the early stages in patients who are in rehab, and thus relapse can be prevented. Data such as heart rate and heart rate variations, galvanic skin response that is closely related to stress levels, skin temperature in addition to the GPS location of the patient are collected. Processed by Machine learning algorithms, this data will provide insights into the behavioral and physiological changes the patient is undergoing which indicate if the patient is in a pre-relapse craving state. Those who are in this zone have a higher propensity to relapse. Alerts can be sent to healthcare providers to take proactive measures targeted to these patients to prevent a relapse.
Real-time examples of the success of technology in battling the opioid crisis:
- The Measure Action Prevent (MAP) Health Management was formed in 2011 in response to the failing efforts in treating substance abuse patients. MAP’s CEO, Jacob Levenson1 calls the opioid crisis as a chronic illness and that the US government tries to fight it using an acute care model which is never going to work. He calls for the patients to be more aware of the risks of addiction and fend for themselves. MAP has relied on predictive analysis and ML-based inputs along with cognitive computing for early detection. MAP has been using AI technology for enhancing their services by exploiting its capability to organize and structure the large amounts of unstructured data in their possession and improve the way in which they leverage patient data to understand the risk of relapse. Using tools such as Telehealth, remote Bluetooth enabled breathalyzers, and biometric pill dispensation kiosks, MAP has an agreement with over 200 healthcare providers in the US to improve behavioral health and addiction treatment for substance abuse patients. With quality data flowing in and the analytical capacity of AIs, MAP hopes to accurately tailor treatment to individual needs.
Levenson1 also says that he sees no end in sight to the opioid crisis and data-informed treatment is the only way in which it can be tackled effectively.
- The USC Machine Learning Center has implemented Machine Learning algorithms to detect opioid dependency. The center has trained its algorithms on one of the largest data sets comprising of 102,166 patients. The PHI was used in accordance with patient privacy and data use consent standards. The center collaborated with the Mayo Health Clinic, a provider known for its partnership with pioneering technological companies for the sake of the betterment of healthcare. Results from the ML-based data analysis showed that among the surveyed patients 79% were short-term users of opioids and 21% were long-term users. A small percentage of 3.47% were categorized as opioid-dependent. Another, even smaller percentage of 0.7% were categorized as opioid-dependent even before the study was conducted. Through this research USC targets to use PHI data as the cornerstone for machine learning algorithms so that opioid dependency can be identified before a physician ever prescribes opioids to patients who are already at high risk of developing a dependency. The center’s director says that the machine learning studies are part of an ongoing study in an effort to bring tools to the market that can effectively combat opioid addiction.
Author: Mohan Bharathi