Study Uses Innovative Technology to Predict Second HFx and Minimize Risks in Elderly Patients
A recent study conducted at the Hospital Universitario Infanta Leonor in Madrid has utilized innovative technology to predict the occurrence of a second hip fracture (HFx) in elderly patients and develop strategies to minimize the associated risks. The study, which analyzed real-world data collected from January 2011 to December 2019, aimed to characterize patients who experienced a second HFx and develop a predictive model for such fractures.
The researchers employed EHRead®, a technology that combines natural language processing (NLP) and machine learning (ML), to analyze the free-text information in electronic health records (EHRs). Additionally, statistical techniques were utilized to compare the patients and develop a predictive model for the occurrence of a second HFx.
The findings of the study will play a crucial role in defining the target population that would benefit from multimodal treatment strategies to minimize the occurrence of a second HFx in at-risk elderly patients.
The retrospective cohort study included patients aged 65 years or older who suffered from HFx during the study period. The index or inclusion date was defined as the first mention of HFx in the SP. Patients were considered to have a second HFx if a contralateral HFx was detected during the study period. Several exclusion criteria were applied, such as the absence of fracture laterality, less than one-month follow-up, detection of a contralateral HFx within the first month after the initial fracture, and the presence of a previous HFx before the index date.
All emergency, inpatient, outpatient records, and hospital pharmacy and laboratory data were used to characterize the patients and events of interest both at baseline and during the follow-up period.
Conceptual definitions of the study variables were pre-specified and mapped to clinical entities present in SNOMED CT, a comprehensive collection of medical terms used in clinical documentation. The accuracy of the conceptual definitions and entity mapping was reviewed and approved by two specialists.
EHRead® was then employed to extract the mapped clinical entities from the EHRs. This proprietary technology utilizes NLP and ML to extract clinical entities and their context from free text.
Variables were constructed by applying specific data wrangling operations to the mapped entities, taking into account NLP parameters, metadata, and record-specific information. Socio-demographic characteristics, clinical characteristics, comorbidities, medication use, disability measures, and other relevant parameters were considered for the analysis.
Descriptive analyses of baseline and follow-up characteristics were performed separately for patients who experienced a second HFx and those who did not. Furthermore, a detailed analysis of mortality corresponding to the median time until refracture (1.3 years) was conducted to understand the potential influence of death as a competitive event.
A subdistribution hazard (Fine-Gray) competing-risk model was developed to predict the occurrence of a second HFx, considering death as the competitive event. The model development process involved selecting an appropriate follow-up duration, determining the maximum number of predictors based on the sample size and event rate, selecting predictors based on clinical criteria, and applying backward feature selection to refine the model.
To ensure patient confidentiality, data were collected from pseudonymized EHRs, and only study-specific variables were extracted. The study adhered to ethical and regulatory standards for good research practices.
In conclusion, this study utilized innovative technology to predict the occurrence of a second hip fracture in elderly patients and develop strategies to minimize associated risks. The findings will help identify the target population for implementing multimodal treatment strategies, ultimately reducing the occurrence of second hip fractures in at-risk individuals.