{"doi":"10.1111/jgs.16831","title":"RESEARCHComparing Strategies for Identifying Falls in Older Adult Emergency Department Visits Using <scp>EHR</scp> Data","abstract":"Emergency department (ED) visits for falls among older adults are often sentinel events for poor health trajectories; however, challenges exist in defining fall-related visits in the ED. Diagnosis codes are a standard method for identifying falls in large administrative data sets. Although a common practice for many conditions, this strategy may miss many patients presenting with falls in the ED. Natural language processing (NLP) refers to a set of techniques by which language as written or spoken by humans can be rendered analyzable by computation. Our team recently developed and validated a simple rules-based NLP system that accurately identified falls from the text of ED physician notes.1 Although this performance is encouraging, barriers exist to using this methodology for all investigations of falls. Researchers, administrators, and epidemiologists often face tradeoffs between large data sets that cover populations of interest but do not have granular clinical data and data sets that are clinically generated and contain more information on fewer patients. Insurance claims, for instance, generate comprehensive data on large populations, making claims-based data sets attractive for studying fall epidemiology. However, they do not contain provider notes or other clinical information. Furthermore, even when text is available in clinical data sets, application of NLP requires programming expertise for retrospective studies and potentially new information technology infrastructure at the health system level to process large numbers of notes or real-time implementation within an electronic health record (EHR).2, 3 Given these barriers to applying NLP, claims-based strategies will continue to have a role in studying falls. When deciding on and applying a strategy for identifying falls, it is critical to understand relevant performance characteristics to interpret results or design interventions. The goal of this study was to compare performance characteristics of several fall identification strategies using EHR data from ED visits using manual chart abstraction as a gold standard. We performed a retrospective observational study using EHR data at an academic medical center with approximately 60,000 ED visits per year. The study was approved by the institutional review board. The current analysis uses the same set of 500 ED visits from unique patients aged 65 and older as did the previous NLP validation study.1 Visits were randomly selected from the period between December 13, 2016, and April 24, 2017. International Classification of Diseases (ICD) codes: Two previously described fall identification strategies based on the ICD code were tested.4, 5 These strategies were developed using ICD-9 codes (listed below). We created a crosswalk to migrate our ICD-10 data from the more limited ICD-9 codes and then applied the ICD-9 fall definitions. Each method was compared with manual abstraction, using Stata v.15 (StataCorp, College Station, TX) and MedCalc (MedCalc Software, Ostend, Belgium) to calculate statistics. Accuracy, specificity, sensitivity, positive predictive value (PPV), negative predictive value (NPV), likelihood ratios, and F1 scores were calculated for each comparison. The F1 score refers to the harmonic mean of PPV and sensitivity, and it is often used as a measure of accuracy in the field of document retrieval. We used 95% confidence intervals to determine significant differences between fall identification strategies. Of the 500 ED visits, 494 were able to be matched with abstracted hospital administrative data. Human coders (manual abstraction) determined that 119 of the 494 ED provider notes (24.1%) explicitly mentioned a fall as a contributing reason for that visit. All five code-based strategies identified fewer falls, ranging from 72 (14.6%) identified using CC to 116 falls (23.5%) identified using the Broad/CC strategy. Accuracy, sensitivity, specificity, PPV, NPV, and F1 score are presented in Table 1. When comp","journal":"Journal of the American Geriatrics Society","year":2020,"id":83659,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":11,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9388,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2020-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":430260,"name":"Gwen Costa Jacobsohn","orcid":"0000-0003-0625-3589","position":1,"is_corresponding":false},{"id":430261,"name":"Apoorva Maru","orcid":"0000-0001-6184-6446","position":2,"is_corresponding":false},{"id":413443,"name":"Arjun K. Venkatesh","orcid":"0000-0002-8248-0567","position":3,"is_corresponding":false},{"id":430262,"name":"Maureen A. Smith","orcid":"0000-0003-4370-000X","position":4,"is_corresponding":false},{"id":430263,"name":"Manish N. Shah","orcid":"0000-0001-6331-1074","position":5,"is_corresponding":false},{"id":111245,"name":"Eneida A. Mendonca","orcid":"0000-0003-4297-9221","position":6,"is_corresponding":false},{"id":430259,"name":"Brian W. Patterson","orcid":"0000-0002-4584-3808","position":0,"is_corresponding":true}],"reference_count":12,"raw_metadata":null,"created_at":"2026-07-18T21:54:32.422022Z","pmid":"32951200","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}