NLP-based fall risk prediction in the elderly.

Smart Care for the Elderly: Development and validation of a Natural Language Processing (NLP)-based model for fall risk prediction in the elderly based on free text from the Electronic Patient Record (EPD). - NLP-based fall risk prediction in the elderly

Status

Active, not recruiting

Phases

Unknown

Study type

Observational

Source

NL-OMON

Registry ID

NL-OMON58429

Enrollment

4000

Registered

2026-03-02

Start date

2026-03-15

Completion date

Unknown

Last updated

2026-03-16

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Fall risk prediction in geriatrics Fall risk prediction in geriatrics

Interventions

None listed

Eligibility

Inclusion criteria

Inclusion criteria: Admissions of patients who were admitted to Deventer Hospital's B2 ward for at least 24 hours between January 1, 2021, and December 31, 2025.Admissions where the patient was 70 years or older at the time of admission.

Exclusion criteria

Exclusion criteria: No exclusion criteria

Design outcomes

Primary

Measure	Time frame
The primary outcome measure is the predictive performance of the NLP prediction model for individual fall risk in inpatients aged 70 years and older. The fall risk generated by the model is compared retrospectively with the actual occurrence of a recorded fall during the hospital stay. The primary evaluation measure is the model's discrimination, expressed as the area under the receiver operating curve (AUROC).	—

Measure

Time frame

The primary outcome measure is the predictive performance of the NLP prediction model for individual fall risk in inpatients aged 70 years and older. The fall risk generated by the model is compared retrospectively with the actual occurrence of a recorded fall during the hospital stay. The primary evaluation measure is the model's discrimination, expressed as the area under the receiver operating curve (AUROC).

—

Secondary

Measure	Time frame
The secondary outcomes of the study are as follows:- Additional performance measures of the fall risk model, including:o Sensitivity and specificity;o F1 scoreo Positive and negative predictive value;o Calibration (e.g., via Brier score).- The performance of alternative modeling approaches for fall risk prediction, including:o Models based solely on structured clinical data;o Models based on risk factors extracted from free-form text;o Models based directly on free-form text (black-box approach);o Combinations of the above- The accuracy of the NLP model in recognizing fall-related risk factors in free-form EHR text, expressed in measures such as precision, recall, and F1 score, based on manually annotated text.- The clinical utility and perceived added value of the generated model outcomes (fall risk and risk factors), evaluated among nurses.	—

Measure

Time frame

The secondary outcomes of the study are as follows:- Additional performance measures of the fall risk model, including:o Sensitivity and specificity;o F1 scoreo Positive and negative predictive value;o Calibration (e.g., via Brier score).- The performance of alternative modeling approaches for fall risk prediction, including:o Models based solely on structured clinical data;o Models based on risk factors extracted from free-form text;o Models based directly on free-form text (black-box approach);o Combinations of the above- The accuracy of the NLP model in recognizing fall-related risk factors in free-form EHR text, expressed in measures such as precision, recall, and F1 score, based on manually annotated text.- The clinical utility and perceived added value of the generated model outcomes (fall risk and risk factors), evaluated among nurses.

—

Countries

Netherlands

Contacts

Public ContactJ. Faber

Deventer Ziekenhuis

j.faber@dz.nl0570-536248

Outcome results