Measurement of Fall Injury With Health Care System Data and Assessment of Inclusiveness and Validity of Measurement Models

Abstract

Importance: National injury surveillance systems use administrative data to collect information about severe fall-related trauma and mortality. Measuring milder injuries in ambulatory clinics would improve comprehensive outcomes measurement across the care spectrum.

Objectives: To assess a flexible set of administrative data-only algorithms for health systems to capture a greater breadth of injuries than traditional fall injury surveillance algorithms and to quantify the algorithm inclusiveness and validity associated with expanding to milder injuries.

Design, setting, and participants: In this longitudinal diagnostic study of 13 939 older adults (≥65 years) in the nationally representative Health and Retirement Study, a survey was conducted every 2 years and was linked to hospital, emergency department, postacute skilled nursing home, and outpatient Medicare claims (2000-2012). During each 2-year observation period, participants were considered to have sustained a fall-related injury (FRI) based on a composite reference standard of having either an external cause of injury (E-code) or confirmation by the Health and Retirement Study patient interview. A framework involving 3 algorithms with International Classification of Diseases, Ninth Revision codes that extend FRI identification with administrative data beyond the use of fall-related E-codes was developed: an acute care algorithm (head and face or limb, neck, and trunk injury reported at the hospital or emergency department), a balanced algorithm (all acute care algorithm injuries plus severe nonemergency outpatient injuries), and an inclusive algorithm (almost all injuries). Data were collected from January 1, 1998, through December 31, 2012, and statistical analysis was performed from August 1, 2016, to March 1, 2019.

Main outcomes and measures: Validity, measured as the proportion of potential FRI diagnoses confirmed by the reference standard, and inclusiveness, measured as the proportion of reference-standard FRIs captured by the potential FRI diagnoses.

Results: Of 13 939 participants, 1672 (42.4%) were male, with a mean (SD) age of 77.56 (7.63) years. Among 50 310 observation periods, 9270 potential FRI diagnoses (18.4%) were identified; these were tested against 8621 reference-standard FRIs (17.1%). Compared with the commonly used method of E-coded-only FRIs (2-year incidence, 8.8% [95% CI, 8.6%-9.1%]; inclusion of 51.5% [95% CI, 50.4%-52.5%] of the reference-standard FRIs), FRI inclusion was increased with use of the study framework of algorithms. With the acute care algorithm (2-year incidence, 12.6% [95% CI, 12.4%-12.9%]), validity was prioritized (88.6% [95% CI, 87.4%-89.8%]) over inclusiveness (62.1% [95% CI, 61.1%-63.1%]). The balanced algorithm showed a 2-year incidence of 14.6% (95% CI, 14.3%-14.9%), inclusion of 65.3% (95% CI, 64.3%-66.3%), and validity of 83.2% (95% CI, 81.9%-84.6%). With the inclusive algorithm, the number of potential FRIs increased compared with the E-code-only method (2-year incidence, 17.4% [95% CI, 17.1%-17.8%]; inclusion, 68.4% [95% CI, 67.4%-69.3%]; validity, 75.2% [95% CI, 73.7%-76.6%]).

Conclusions and relevance: The findings suggest that use of algorithms with International Classification of Diseases, Ninth Revision codes may increase inclusion of FRIs by health care systems compared with E-codes and that these algorithms may be used by health systems to evaluate interventions and quality improvement efforts.

Figure 1.. Construction of Fall-Related Injury (FRI) Episodes of Care From Claims

Claims for external cause of injury–coded (E-coded) reference standard or International Classification of Diseases, Ninth Revision (ICD-9)–coded potential FRIs were grouped into episodes of care during 2-year observation windows (the unit of analysis) before the interviews. Left, a potential FRI episode, an initial hip fracture (HF) ICD-9 code, and multiple subsequent same-injury diagnoses for follow-up care within 180 days. Middle, a skull fracture (SF) ICD-9 was excluded because of an E-code for nonfall (motor vehicle crash). Right, a single diagnosis code indicates a potential FRI involving an ankle dislocation (AD).

Figure 2.. Consolidated Categories of Potential Fall-Related Injury (FRI) Episodes by Sample Size and Validity

Prevalence (entire bar) and validity (dark gray) of consolidated groups of potential FRI International Classification of Diseases, Ninth Revision (ICD-9)–coded test episodes before testing in the algorithms. Cases are number of 2-year observation periods containing episodes. Left, FRI episodes involving care in acute settings (hospital, emergency department, and postacute nursing facilities). Right, all inpatient and outpatient settings. Each category is considered independently from the others; the same observation window can be included more than once.

Figure 3.. Inclusiveness vs Validity in 3 Fall-Related Injury (FRI) Administrative Data Algorithms

Cumulative result of adding test International Classification of Diseases, Ninth Revision (ICD-9)–coded injuries to algorithm for inclusiveness and validity. Left of dotted line, E-coded FRIs with additions until all potential ICD-9–coded injuries are included into the test algorithm. Inclusion indicates percentage of reference-standard observation windows (9904 periods) captured by the algorithm up to that point. Validity indicates proportion of test FRIs within 6 months of an interview up to that point (an increasing denominator, from 0 test FRIs until the maximum of 3936 observation periods) validated by the reference standard.