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

This diagnostic study uses Health and Retirement Survey and linked Medicare data to develop and validate an administrative approach to identifying older adults with fall-related injuries and to quantify the inclusiveness and validity across a spectrum of potential diagnoses of fall-related injuries.

Technical notes on grouping International Classification of Disease-9 (ICD-9) diagnostic codes.
Potential Fall-related Injury types identified by ICD-9 codes were organized into a matrix of anatomic regions and injury types. We conceptualized the composite reference standard (E-code for fall or patient-reported fall injury) as similar to a "gold standard" for testing the matrix categories using traditional diagnostic test properties. Of the test properties, we focused on positive predictive value (PPV), or percent of test alternative injuries validated by the composite reference standard, as our criterion to simplify and prioritize the matrix. The second focus of the project was inclusion, or sensitivity (SEN), or the percent of the composite reference standard captured by the matrix categories. First, we collapsed matrix categories low, medium and high accuracy. Low PPV was roughly determined by <70% (gray), medium PPV was 70-79% (blue), high PPV was >80% (orange). For cells less than 20, we grouped together with other categories in the same anatomic region if within one PPV category higher. We grouped the injuries separately using these criteria for acute care (emergency department/hospital/post-hospital nursing home) injuries vs all injuries (acute care plus the outpatient clinic data). Consolidated groups are labeled with group ID number and color coded (orange=high; blue=medium; gray=low). Groups in gray without a group ID number (<50% accuracy) were further refined by specific diagnoses.

General Overview
• eTable 5 is the required first step for all three algorithms.
• If the Acute Care algorithm is desired, then please refer to eTable 5 then 6 • If the Balanced algorithm is desired, please refer to eTable 5 then 7 • If the Inclusive algorithm is desired, please refer to eTable 5 then 8 eTable 5: SAS code to analyze Medicare claims eligible for potential fall injury • Concatenate all the data sources from Medicare inpatient, outpatient (this includes all outpatient clinic and emergency room data) nursing home claims files. • Include all injuries with potential for fall injury (all injuries in ICD-9 800-999 except those types listed below) • Group the injuries into anatomic locations (e.g., head, limb) and type of injury (e.g., fracture, contusion) • Apply inclusions for outpatient clinic data which require confirmation with CPT codes (procedures and imaging) that are specific to anatomic locations (e.g., an arm cast for arm fracture or arm x-ray for arm contusion) • Exclude if E-codes other than fall are found in the claim (e.g., auto accident E-code would exclude a hip fracture claim). • Exclude if advanced cancer or bone metastatic disease is found in the claim eTable 6: SAS code for the fall algorithm 1 (Acute care) • Create episodes of care for all claims within the same category of injury within 180 days of each other. • Create episodes of care by fall-related E codes within 180 days of each other • Exclude any episode of either type that begins with a non-fall E-code • Flag episodes with any E-code for falls plus the 5 consolidated categories of ICD-9 FRI episodes in the Acute Care algorithm. All of these episodes include at least some care in the hospital or emergency room. • The first date of service for each episode is considered to be the date of the fall injury eTable 7: SAS code for the fall algorithm 2 (Balanced) • Create episodes of care for all claims within the same category of injury within 180 days of each other. • Create episodes of care by fall-related E codes within 180 days of each other • Exclude any episode of either type that begins with a non-fall E-code • Flag episodes with any E-code for falls plus the 5 consolidated categories of ICD-9 FRI episodes in the Acute Care algorithm, plus 4 consolidated categories of outpatient clinic FRI episodes that results in the Balanced Algorithm. The first date of service for each episode is considered to be the date of the fall injury eTable 8. SAS code for fall algorithm 3 (Inclusive Algorithm) • Create episodes of care for all claims within the same category of injury within 180 days of each other. • Create episodes of care by fall-related E codes within 180 days of each other • Exclude any episode of either type that begins with a non-fall E-code • Flag episodes with any E-code for falls, the 5 consolidated categories of ICD-9 FRI episodes in the Acute Care algorithm, plus 4 consolidated categories of outpatient clinic FRI episodes in the Balanced Algorithm, plus 2 consolidated categories of outpatient clinic FRI episodes to form the Inclusive Algorithm. The first date of service for each episode is considered to be the date of the fall injury

Technical notes on potential fall-related injury diagnosis codes
In eTable 5, when we collected all claims with injuries that could plausibly be caused by a fall, we considered all International Classification of Diseases-9 injury codes from 800 to 999. Of these, we excluded these types of injuries considered to be highly implausible to be due to a fall: •