Screening medical records using artificial intelligence – the right algorithm in primary care

AI and precision medicine are changing the way we detect and treat diseases [1]. Perhaps the strongest impact is sentiment in primary care, where data-driven decision support tools will aid in the care of patients diagnosed and treated as soon as possible. [2, 3].

The question is how best to design systems to benefit the health care system and patients. Bone arterial disease (BAS) is a common but under-detected condition and therefore untreated. A nationwide study found that just over 1 in 5 patients with BAS in hospital died within a year and 1 in 6 had a major cardiovascular event within a year. [4]. The prevalence of BAS in primary care is high, but physicians’ awareness of the diagnosis is relatively low [5].

Diagnosis of BAS is based on anamnesis, vascular status, and calculation of the so-called ankle-brachial index (ABI). [6]. The cost-effectiveness of ABI screening depends on the prevalence of BAS in a group, treatment cost/month and adherence to treatment [7]. Therefore, screening should focus on high-risk cohorts [8]. Data-based BAS risk indicators can greatly improve screening specificity [9]. By evaluating the risks to individual patients using a data-driven algorithm, the target population can be reduced and the accuracy increased [10].

The accuracy of risk prediction models increases as more data becomes available. Through the Data Protection Regulation (GDPR), an individual is given the right to receive and transfer personal data from one system to another. Health data can therefore be collected, with the consent of the individual, from various sources in order to deliver improved health services based on algorithms.

More data requires carefully designed systems that do not overburden employees with unnecessary alerts and information. Systems need to adopt different strategies depending on increasing levels of risk to make decisions. Suggestions for lifestyle change – for example how to quit smoking or get more exercise – can be generated by AI algorithms in dialogue with patients. Routine activities can be controlled by risk algorithms on demand; For example, by suggesting that people at high risk should be seen and evaluated first.

In order to convert research results into products that support primary care, the following should be considered:

• The development of data-driven models should rely, as far as possible, on health care information that is easily accessible in the various primary care IT systems.
• New medical technology requirements must be placed on software that ensures its use.
• Healthcare providers should formulate clear guidelines and routines for how to use AI-based systems for decision support.

Today, more than 80 percent of Swedish primary care has access to a high-quality follow-up tool that identifies an increased risk of cardiovascular disease in people with diabetes. During the year, patients with an increased risk of BAS . will also be identified [11]. The tool is integrated into the local medical record and provides a preventive method of action. Through the Swedish Municipalities and Regions (SKR) initiative “Primary Care Quality” [12] In addition, about 90 percent of health care centers have access to their own data and follow national quality indicators.

AI-powered log screening should be used to identify high-risk individuals. Better accuracy in primary care means better quality of life for the individual, less risky health outcomes and significant savings. The widespread use of data-driven decision support will change a lot quickly, not least for those residents who can receive rapid diagnosis and treatment early in the disease course. Care is burdened with fewer serious events that require inpatient care and a greater impact than existing treatments.

In collaboration with specialists in data-driven modeling, Swedish healthcare has the regulatory and competency requirements to lead development.

Medical Journal 41-42 / 2022

Lakartidningen.se 2022-10-13