NLP technology identifies social determinants of health (SDOH) for Alzheimer’s patients using unstructured EHR data, aiding clinicians in addressing transportation, housing, isolation, and more. A study developed a rule-based NLP algorithm, that outperforms logistic regression and deep learning models. Prospective validation against an SDOH questionnaire aims to refine the algorithm’s utility in enhancing outcomes for dementia patients by connecting them with community resources.
A recent study published in Health Services Research highlights the application of natural language processing (NLP) technology to identify crucial social determinants of health (SDOH) for individuals suffering from Alzheimer’s disease and related dementias (ADRD), utilizing unstructured electronic health records (EHR) data.
Researchers have devised an NLP algorithm capable of pinpointing SDOH factors within the context of ADRD patients, leveraging unstructured EHR data. The algorithm is proficient at detecting SDOH elements like transportation, nutrition, housing, financial hardships, social isolation, instances of abuse or neglect, and concerns regarding medication access. The research emphasizes that these SDOH factors are pivotal determinants influencing adverse health outcomes in ADRD patients.
A notable challenge arises from the fact that many healthcare settings lack a systematic approach to collecting essential SDOH information. However, this information can often be gleaned from unstructured EHR data, making access and utilization complex.
The researchers propose that NLP holds the potential to effectively extract and utilize this information within clinical contexts. By automating the extraction process, the NLP tool could support clinicians, social workers, and case managers in recognizing patients’ SDOH needs and implementing preventive measures against adverse events.
The development of the algorithm involved analyzing 1,000 randomly selected medical notes from 7,401 inpatient social worker and emergency department reports created between 2015 and 2019 for 231 ADRD patients at Michigan Medicine, the University of Michigan’s academic medical center.
Based on this data, the research team designed a rule-based NLP algorithm and trained it to identify the seven categories of social determinants mentioned earlier.
The comparative analysis included the NLP model, a regularized logistic regression approach, and a deep learning model, all trained to achieve the same objective. The evaluation was based on metrics like accuracy, sensitivity, specificity, F1 score, and the area under the receiver operating characteristic curve.
Each tool employed 700 training notes, with the remaining 300 reserved for validation purposes. The NLP algorithm exhibited F1 and AUC scores of at least 0.94 and 0.95 respectively across all SDOH categories using the 700 training notes. With the 300 validation notes, the algorithm achieved F1 and AUC scores of at least 0.80 and 0.97 for all SDOH domains, excluding housing and medication insecurities.
In comparison, the regularized logistic regression and deep learning models demonstrated lower performance, underscoring the NLP model’s significant superiority.
These results suggest the potential for NLP models to effectively flag SDOH factors, thereby enhancing patient outcomes. However, further research is required to validate these findings.
Moving forward, the research team intends to validate the NLP model against an SDOH questionnaire currently implemented in Michigan Medicine’s primary care settings. This validation process aims to compare the algorithm’s outputs with patient responses, potentially refining its performance.
Elham Mahmoudi, Ph.D., a health economist at Michigan Medicine and a lead researcher on the project, shared, “We are also preparing a pilot program that will evaluate the feasibility of an intervention that addresses these social determinants of health, and connect identified people with community resources.” She added, “In the meantime, we hope our current results show that this algorithm can be used by clinicians, case managers, and social workers to proactively address the social needs of patients with dementia, and potentially other vulnerable patient populations.”
