Abstract
Background
Fall-related injuries (FRIs) are a leading cause of morbidity, mortality, and costs among nursing home (NH) residents. Carefully defining FRIs in administrative data is essential for improving injury-reduction efforts. We developed a series of novel claims-based algorithms for identifying FRIs in long-stay NH residents.
Methods
This is a retrospective cohort of residents of NH residing there for at least 100 days who were continuously enrolled in Medicare Parts A and B in 2016. FRIs were identified using 4 claims-based case-qualifying (CQ) definitions (Inpatient [CQ1], Outpatient and Provider with Procedure [CQ2], Outpatient and Provider with Fall [CQ3], or Inpatient or Outpatient and Provider with Fall [CQ4]). Correlation was calculated using phi correlation coefficients.
Results
Of 153 220 residents (mean [SD] age 81.2 [12.1], 68.0% female), we identified 10 104 with at least one FRI according to one or more CQ definition. Among 2 950 residents with hip fractures, 1 852 (62.8%) were identified by all algorithms. Algorithm CQ4 (n = 326–2 775) identified more FRIs across all injuries while CQ1 identified less (n = 21–2 320). CQ2 identified more intracranial bleeds (1 028 vs 448) than CQ1. For nonfracture categories, few FRIs were identified using CQ1 (n = 20–488). Of the 2 320 residents with hip fractures identified by CQ1, 2 145 (92.5%) had external cause of injury codes. All algorithms were strongly correlated, with phi coefficients ranging from 0.82 to 0.99.
Conclusions
Claims-based algorithms applied to outpatient and provider claims identify more nonfracture FRIs. When identifying risk factors, stakeholders should select the algorithm(s) suitable for the FRI and study purpose.
Keywords: Accidental falls, Algorithms, Fractures, Health care administrative claims, Nursing home
Thirty percent of older adults living in the community and more than half of residents of nursing homes (NHs) will fall each year (1–4). Fall-related injuries (FRIs) are the most common source of injury among older adults and result in approximately 6% of total Medicare spending annually (2,5). FRIs are associated with a wide array of subsequent adverse health outcomes including disability, functional impairment, anxiety, social isolation, cognitive decline, and depression (1,6,7). Additionally, FRIs may result in mortality: in 2015, over 28 000 fall-related deaths were identified in US older adults. From 2004 to 2017, the rate of fatal fall injuries increased 35% (8). The risk of 1-year mortality following a hip fracture, the most common fall-related fracture to result in hospitalization, is approximately 20% in community dwellers and 40% in NH residents (7,9,10). Falls can also result in fractures of other bones, dislocations, concussions, and intracranial bleeds, all of which can contribute to excess morbidity and mortality in this population. Identifying FRIs facilitates the study of risk factors for these injuries as well as the opportunity to test intervention programs for older adults at risk. One key challenge with FRIs, however, is a comprehensive measurement to facilitate these studies.
Linking administrative claims data with other data sources offers a practical approach for studying FRIs on a large scale. Multiple algorithms with high positive predictive values (PPVs) have been developed to identify fractures or FRIs using inpatient claims codes from the International Classification of Diseases (ICD), 9th revision and external cause of injury codes (historically called “E-codes”) (11–16). However, rules from the US Department of Health and Human Services mandated that, starting October 1, 2015, Medicare claims must be submitted using codes from the 10th revision of the ICD (ICD-10) (17). This revision made substantial changes including the addition of thousands of new codes and a shift in the use and designations of external cause of injury codes.
Even prior to the recent revision to ICD-10 there were no administrative FRI algorithms developed specifically for use with the NH population, where nonoperative management of hip fracture occurs twice as frequently as in the general Medicare population (7,18,19). Differences in treatment for FRIs among NH residents may shift the setting in which care is received and change how providers and treatment facilities bill Medicare for services rendered. Thus, a novel means for FRI identification among NH residents utilizing the recently implemented ICD-10 coding system is needed. Our objective was to develop and compare a series of algorithms that could be used by researchers, policymakers, and other stakeholders to identify FRIs using Medicare claims data for US residents living in NHs.
Method
Study Design and Data Sources
This study was a retrospective cohort of residents of NH identified from Medicare administrative claims data using Minimum Data Set v3.0 (MDS) clinical assessment data. Eligibility characteristics were obtained from the Medicare Master Beneficiary Summary File. We included adjudicated claims submitted for inpatient services (Medicare Provider Analysis and Review or MedPAR) and outpatient services (both Outpatient research identifiable file [RIF] and Carrier RIF) for the calendar year 2016. MedPAR claims included short-stay inpatient, long-stay inpatient, and skilled nursing facility claims. Claims submitted by noninstitutional providers, such as ambulance companies, were included in the Carrier file. External cause of injury codes on the same date as the diagnostic codes that were submitted to Medicare and included in MedPAR, Outpatient RIF, or Carrier RIF were included. Physician’s Current Procedural Terminology, fourth edition (CPT-4) procedural claims are unique to Carrier file claims and were considered in the Outpatient and Provider with Procedure algorithm (CQ2). We utilized a 20% sample of the Carrier file.
For each resident, Medicare claims were linked with the MDS records. MDS data arise from a clinical assessment with more than 300 items, completion of which is federally required of all NHs receiving Medicare or Medicaid funding (20–22). The MDS includes assessments for all residents at admission and discharge in addition to periodic evaluations conducted quarterly, annually, and upon any significant change in health status. MDS data were included from October 1, 2015 through December 31, 2016.
Study Population
Supplementary Figure 1 describes the selection of participants. The retrospective cohort included all residents of NHs 65 years or older. Residents with a prolonged stay were defined as those who spent at least 100 days in an NH facility with no more than 10 consecutive days outside the facility (3,23). A resident’s index date was defined as the day the resident qualified as long-stay. For residents who qualified as long-stay prior to 2016, an index date of January 1, 2016 was applied. To be included in this analysis, residents must have achieved long-stay status no later than January 31, 2016. We included residents continuously enrolled in Medicare fee-for-service Parts A and B with at least 1 Part A claim and 1 Part B claim through December 31, 2016. Because the MedPAR file does not contain complete FRI information for all Medicare Advantage enrollees, we excluded residents enrolled in a Medicare Advantage plan. We further excluded residents without any MDS assessment in the year prior to the index date (20). Finally, we restricted our sample to the approximately 20% sample of residents for whom we had available Carrier claims in 2016. Our final sample included 153 220 NH residents.
Identifying FRIs Through Claims-Based Algorithms
Because there is no gold standard approach to identify FRIs using Medicare claims data, we developed a series of case-qualifying (CQ) definitions to detect FRIs, similar to the approach used by Wright et al. (14) to identify all fractures excluding the hands and feet. FRIs were ascertained between January 2016 and December 2016. Not all FRIs require surgical intervention; as many as 11.8% of NH residents with hip fracture do not undergo surgical repair, twice the rate seen in the overall Medicare population (7,18). As such, we developed algorithms that could broadly capture FRIs that may require a spectrum of services ranging from inpatient stays with surgical intervention to outpatient management (19,24). We previously demonstrated that the inclusion of Medicare Part B increases the number of hip fracture events identified in residents of NHs (24). Furthermore, outpatient management may be more likely in areas with differences in access to inpatient or specialty care along with regional differences in management (25).
Table 1 presents the overview of the algorithms created using Medicare claims data. These algorithms are not intended to be hierarchical; each algorithm is distinct and has its own potential application based on the type of injury, practice setting, and necessity of ensuring that the injury occurred in the setting of a fall. The Inpatient algorithm (CQ1) requires an inpatient diagnostic claim for an FRI and is intended to identify patients who were admitted to the hospital with an FRI. The Outpatient and Provider with Procedure algorithm (CQ2) requires a Carrier file diagnostic claim for an FRI with an associated CPT-4 procedural code for repair, closed reduction, casting, strapping, or splinting related to the injury type that appears on the same encounter. We allowed for some overlap in the confirmatory procedural codes between injury sites (eg, hip fracture with repair of distal femur). The CQ2 algorithm is intended to identify both hospitalized and nonhospitalized injuries, including injuries that were managed nonoperatively. Importantly, neither the CQ1 nor CQ2 algorithms require an external cause of injury code indicative of a fall. In contrast, the Outpatient and Provider with Fall algorithm (CQ3) requires a Carrier file diagnostic claim for an FRI plus either an ICD-10 code or external cause of injury code associated with a fall on the same date (Supplementary Table 1). We excluded external cause of injury codes related to traumatic events that may not be due to a fall, such as transport accidents or assault (Supplementary Table 2). The CQ3 algorithm is intended to identify patients who received medical attention in the setting of a fall, with less certainty as to the precise injury experienced (eg, hip pain without a confirmatory procedural code for hip fracture). Finally, the Inpatient or Outpatient and Provider with Fall (CQ4) algorithm identifies a subset of FRIs from CQ1 and CQ2 algorithms with the specification that a fall occurred at the time of the injury plus all injuries identified using the CQ3 algorithm. Specifically, the CQ4 algorithm requires either an inpatient OR carrier file diagnostic claims code AND an ICD-10 external cause of injury code associated with a fall.
Table 1.
Overview of the 4 Different Case-Qualifying Definition Algorithms Used to Identify Fall-Related Injuries Using Medicare Claims Data
| Inpatient Claims (CQ1) | Outpatient and Provider Claims With Procedure (CQ2) | Outpatient and Provider Claims With Fall (CQ3) | Inpatient or Outpatient and Provider Claims With Fall (CQ4) |
|---|---|---|---|
| Inpatient ICD-10 Injury Code | Outpatient or Provider ICD-10 Injury Code AND CPT-4 Procedure Code |
Outpatient or Provider ICD-10 Injury Code AND External Cause of Injury Code |
(Inpatient ICD-10 Injury Code OR Outpatient or Provider ICD-10 Injury Code) AND External Cause of Injury Code |
| Exclude any claim with a code for transport accidents (eg, V00.0), suicide (eg, T14.91), thermal injury (eg, T20.1), poisoning (eg, T50.901A), accidents in nature (eg, W14), adverse medical event (eg, T43.0), assault (eg, Y04.0), or exposure to forces of nature (eg, X31). A full list of excluded codes can be found in Supplementary Table 2. |
Notes: CPT-4 = Current Procedural Terminology, fourth edition; ICD-10 = International Statistical Classification of Diseases and Related Health Problems, 10th revision. Inpatient claims are obtained from the MedPAR data set. Outpatient and provider claims are obtained from Carrier file. Procedure codes are obtained from the Carrier file. Inpatient or Outpatient and Provider claims with Fall (CQ4) may be obtained from either the MedPAR or Carrier file.
Our decision to employ 4 separate CQ definitions was due to the potential tradeoffs inherent in each algorithm. For example, requiring an external cause of injury code ensures the injury was due to a fall, but it may lead to decreased recognition in certain geographic locations because regional mandates to use these codes vary (26). It is also important to consider the type of injury when selecting an algorithm: Several of the included injuries may occur outside of a fall, even in NH residents. Hip fractures have a high likelihood of being fall-related while axial injuries (if not accompanied by an external cause of injury code) have a low likelihood. The creation of multiple algorithms allows researchers to select an algorithm that best matches their research question, including the importance of confirming injury type, the mechanism of injury, and treatment setting. When employing the Outpatient and Provider with Procedure algorithm (CQ2) to identify an FRI, there are a few special injury types where imaging codes, rather than procedural codes, were used to confirm the injury. For example, for vertebral fracture, we considered CPT-4 imaging codes for thoracic and lumbar fractures in the 10 days preceding the diagnostic claim, rather than requiring procedural codes on the same day. This is consistent with prior literature that demonstrates thoracic and lumbar vertebral fracture detection is improved when using imaging codes in the days preceding the diagnostic code (14). We used a similar approach for intracranial bleeding. We did not require confirmatory imaging or a procedural code for concussion or other injuries, as these are clinical diagnoses.
Types of FRIs
We defined an FRI as any injury commonly associated with a fall among NH residents. These included fractures (except toes and fingers), dislocations, concussions, intracranial hemorrhages, or other organ trauma (eg, pneumothorax). We then identified all ICD-10 claims from the ICD-10-CM Tabular List of Diseases and Injuries and CPT-4 repair codes associated with these injuries. While claims codes were updated in 2016, procedure codes were not; we reviewed algorithms published using ICD-9 claims codes to identify associated CPT-4 procedural codes used to confirm FRIs (11,12,14). Diagnostic and procedural codes were reviewed by an expert panel, which consisted of an emergency medicine physician, an orthopedist, and a geriatrician. After review, we included 148 ICD-10 diagnostic codes and 395 CPT-4 procedural codes in our algorithm. FRIs were grouped using the following categories: hip fracture, other lower extremity fracture (distal femur, patella, tibia/fibula, ankle, pelvic), upper extremity fracture (humerus, radius, ulna, carpal), axial fracture (rib, sternum, clavicle, scapula, cervical, lumbar, and thoracic vertebrae), skull/face fracture, dislocations (including vertebral), intracranial bleeding, concussion, and other injury. Supplementary Table 3 provides the full list of ICD-10 claims, grouped by injury type, used in the algorithms. For each injury group, all 4 CQ definitions were applied, yielding 4 separate ways to determine the injury from the claims. Minor injuries (eg, lacerations, contusions, sprains, strains) were excluded because there was substantial overlap with major injuries and these injuries seldom have procedural codes. We provide a list of diagnostic codes associated with minor injuries (Supplementary Table 4).
For all 4 CQ algorithms, we required the diagnostic code for the FRI to be in the primary or first secondary diagnostic position. Furthermore, ICD-10 uses a letter identifier in the last claim position to indicate if a claim is an initial encounter or a sequela. Letter identifiers A, B, or C indicate an initial encounter, and these claims were applied to our algorithm. All claims with other letter identifiers were excluded. For CQ3 and CQ4, we allowed external cause of injury codes in any ICD-10 position to justify inclusion.
Cohort Characteristics
Clinical characteristics were described using the MDS v3.0 assessment closest to and preceding long-stay qualification date. Age and race were measured using the Medicare Master Beneficiary Summary File on the long-stay index date. Activities of daily living (ADLs) were summarized using the MDS ADL-Long Form method, with scores ranging from 0 to 28 with higher scores indicating more impairment (27,28). Cognitive impairment was assessed using the MDS Cognitive Function Scale (21).
Statistical Analysis
Resident characteristics were described according to whether a resident experienced any FRI during follow-up, and then by which CQ definition was used to determine an FRI occurred. We described the frequencies of FRIs by type using each of the 4 algorithms. Each injury type was counted once per resident, but residents could have more than one injury type on the same or different days. To draw inference to years other than 2016, we calculated Clopper–Pearson binomial proportion confidence intervals (CIs) for all injury frequencies, considering 2016 as a sample of all possible years. For the most common and impactful type of FRI, hip fracture, we examined overlap in the number of fractures identified using the 4 different algorithms as a secondary analysis to demonstrate the potential utility of each in different care settings and to illustrate events that would be missed by selecting a single algorithm. We chose hip fracture as the most common FRI in the NH and because it almost always occurs as the result of a fall. We also evaluated the number of additional hip fractures that would be detected if we expanded to the sixth diagnostic position. To understand the extent of agreement between the CQ definitions for hip fracture, we calculated phi correlation coefficients (29,30) (φ) and Cohen’s kappa coefficients (31,32) (κ) along with corresponding 95% CIs. Both measures range from −1 to 1; negative values indicate disagreement, 0 represents independence (φ) or agreement due to random chance (κ), and positive values denote agreement (29–32). We compared the overlap in the frequency of hip fractures identified using inpatient and outpatient claims and with the external cause of injury codes.
Results
Cohort Characteristics
Among the 153 220 NH residents, the mean (SD) age was 81.2 (12.1) years and 68.0% were female. About 40 935 (26.7%) died during the calendar year and 919 (0.6%) were transferred/discharged. By December 31, 2016, 10 104 residents (6.6%) experienced at least one FRI based on one or more of the CQ definitions. The most common type of FRI was hip fracture (n = 2 950), followed by nonhip lower extremity fractures (n = 2 856), axial fractures (n = 2 051), upper extremity fractures (n = 1 488), and intracranial bleeding (n = 1 169). Table 2 describes the clinical characteristics of NH residents according to whether they experienced an FRI. NH residents with one or more FRIs were older (82.3 vs 81.2 years). A greater proportion of residents with at least one FRI had moderate to severe cognitive impairment (42.6% vs 46.7%), had lower ADL scores indicating more independence in performance (15.0 vs 17.0), had a history of falls (66.5% vs 48.8%), and had wandering behavior reported (8.8% vs 4.8%), compared with residents without an FRI. Supplementary Table 5 describes the clinical characteristics of NH residents according to their hip fracture status categorized by each of the 4 CQ definitions. Characteristics were similar between each of the 4 definitions.
Table 2.
Characteristics of a National Cohort of US Residents of Nursing Home With Prolonged Stays Overall and Stratified by the Occurrence of an FRI Between January 1, 2016 and December 31, 2016
| All Residents (n = 153 220) | Residents with FRI* (n = 10 104) | Residents without FRI (n = 143 116) | |
|---|---|---|---|
| Age in years, mean ± SD | 81.2 ± 12.1 | 82.3 ± 11.2 | 81.2 ± 12.2 |
| Female gender, n (%) | 104 248 (68.0%) | 7 486 (74.1%) | 96 762 (67.6%) |
| Race, n (%) | |||
| Non-Hispanic White | 121 756 (79.5%) | 8 685 (86.0%) | 113 071 (79.0%) |
| Non-Hispanic Black/African American | 19 352 (12.6%) | 668 (6.6%) | 18 684 (13.1%) |
| Hispanic | 7 710 (5.0%) | 511 (5.1%) | 7 199 (5.0%) |
| Other/Unknown | 4 402 (2.9%) | 240 (2.4%) | 4 162 (2.9%) |
| Moderate to severe cognitive impairment (CFS ≥3), n (%) | 71 096 (46.4%) | 4 309 (42.6%) | 66 787 (46.7%) |
| Activities of Daily Living Scale, mean ± SD† | 16.9 ± 6.7 | 15.0 ± 6.4 | 17.0 ± 6.7 |
| Fall documented on last assessment, n (%) | 79 230 (51.7%) | 6 723 (66.5%) | 69 907 (48.8%) |
| Always continent (urinary), n (%) | 28 739 (18.8%) | 2 554 (25.3%) | 26 185 (18.3%) |
| Always continent (bowel), n (%) | 54 557 (35.6%) | 4 753 (47.0%) | 49 804 (34.8%) |
| Independent locomotion on unit, n (%) | 21 144 (13.8%) | 1 608 (15.9%) | 19 536 (13.7%) |
| Independent transfer performance, n (%) | 14 017 (9.2%) | 1 158 (11.5%) | 12 859 (9.0%) |
| Wandering, n (%) | 7 788 (5.1%) | 889 (8.8%) | 6 899 (4.8%) |
| Osteoporosis, n (%) | 22 964 (15.0%) | 1 754 (17.4%) | 21 210 (14.8%) |
| Pressure ulcer ≥grade I, n (%) | 9 947 (6.5%) | 531 (5.3%) | 9 416 (6.6%) |
| Diabetes mellitus, n (%) | 50 524 (33.0%) | 3 138 (31.1%) | 47 386 (33.1%) |
Notes: CFS = Cognitive Function Scale; FRI = fall-related injury; SD = standard deviation. Age was measured using Medicare Beneficiary Summary File on the long-stay index date. All other covariates were measured on the MDS evaluation immediately preceding the long-stay index date.
*Fall-related injury as classified by any of the case-qualifying definitions (CQ1–CQ4).
†Score according to the MDS ADL-Long Form Assessment (range 0–28). Higher values indicate greater dependence (less independence) in performing any of the 7 activities of daily living assessed.
Frequency of FRIs
Figure 1 describes the frequencies of FRIs according to the algorithm used. In general, the Inpatient algorithm (CQ1) identified a lower number of events and the Inpatient or Outpatient and Provider with Fall algorithm (CQ4) identified a higher number. More hip fractures were identified using the Inpatient or Outpatient and Provider with Fall algorithm (CQ4; n = 2 775), whereas fewer events were identified by the Outpatient and Provider with Procedure algorithm (CQ2; n = 2 185). For fractures other than the hip, fewer events were identified using the Inpatient algorithm (CQ1) than any of the other CQ definitions. The next most common fractures were nonhip lower extremity fractures (n = 714–2 045), axial fractures (n = 647–2 274), and upper extremity fractures (n = 230–1 429), respectively. The number of FRIs identified varied significantly depending on the algorithm used. Intracranial bleeding (n = 488–1 078), concussions (n = 21–465), and organ trauma (n = 28–550) were identified more using the Outpatient and Provider with Procedure algorithm (CQ2). For nonfracture injury categories, few FRIs were identified using the Inpatient algorithm (CQ1; n = 20–488).
Figure 1.
Frequencies of different types of fall-related injuries among 2016 residents of nursing homes with prolonged stays according to the algorithm used. CQ1 = Inpatient Claims; CQ2 = Outpatient and Provider Claims with Procedure; CQ3 = Outpatient and Provider Claims with Fall; CQ4 = Inpatient or Outpatient and Provider Claims with Fall. 95% confidence intervals were calculated using the Clopper–Pearson method for each prevalence. Some injury categories have been excluded from the figure and suppressed in accordance with the Center for Medicare and Medicaid Services Small Cell Size Suppression policy. Lower extremity fracture includes fractures of the distal femur, patella, tibia, fibula, ankle, and pelvis. Axial fracture includes fractures of the rib, sternum, clavicle, scapula, and vertebrae (including cervical, thoracic, and lumbar). Upper extremity fracture includes fractures of the humerus, radius, ulna, and carpals. Dislocations include dislocation of any joint including vertebral dislocation.
Comparison of Algorithms for Hip Fractures
Figure 2 presents the overlap between hip fractures identified using each of the CQ algorithms. There were 1 852 hip fractures detected by all 4 diagnostic algorithms. Of 2 320, 1 967 (84.8%) of hip fractures identified using the Inpatient algorithm (CQ1) were also identified using the Outpatient and Provider with Procedure algorithm (CQ2). Of 2 185, 218 (10.0%) of hip fractures identified using the Outpatient and Provider with Procedure algorithm (CQ2) were not identified using the Inpatient algorithm (CQ1). An additional 480 hip fractures identified by neither the Inpatient algorithm (CQ1) nor the Outpatient and Provider with Procedure algorithm (CQ2) were identified using the Outpatient and Provider with Fall algorithm (CQ3). As compared with the Inpatient algorithm (CQ1), the Inpatient or Outpatient and Provider with Fall algorithm (CQ4) detected an additional 630 hip fractures. Of 2 320, 2 145 (92.5%) of hip fractures identified using the Inpatient algorithm (CQ1) and 2 018/2 208 (91.4%) of hip fractures identified using the Outpatient and Provider with Procedure algorithm (CQ2) had an associated external cause of injury code for falls. Expanding to the sixth diagnostic position resulted in approximately 100 additional hip fractures (~4% increase) identified in the CQ1 and CQ3 algorithms (Supplementary Table 6) with only 30 additional hip fractures (1.4%) detected in CQ2.
Figure 2.
Venn diagram showing the overlap between hip fracture case-qualifying definitions in Medicare claims among long-stay nursing home residents. If a resident had 2 or more hip fractures during the 1-year follow-up period, only the first was included. CQ1 = Inpatient Claims; CQ2 = Outpatient and Provider Claims with Procedure; CQ3 = Outpatient and Provider Claims with Fall; CQ4 = Inpatient or Outpatient and Provider Claims with Fall.
Agreement Between FRI Algorithms
Evaluation of phi correlation coefficients, as displayed in Table 3, shows strong positive associations between all 4 algorithms, with the weakest correlation between the Outpatient and Provider with Procedure algorithm (CQ2) and the Outpatient and Provider with Fall algorithm (CQ3; φ = 0.82; 95% CI 0.81–0.83). The strongest correlation is between the 2 algorithms (CQ3 and CQ4) requiring external cause of injury codes (φ = 0.99; 95% CI 0.99–0.99). Evaluation of Cohen’s kappa for interrater reliability of the algorithms (Supplementary Table 7) provides similar inferences to that observed using the phi correlation coefficient, with CQ2 and CQ3 having the lowest agreement (κ = 0.81) and CQ3 and CQ4 having the highest (κ = 0.99).
Table 3.
Phi Correlation Coefficients for Hip Fracture Detection by CQ Definition
| CQ1 | CQ2 | CQ3 | CQ4 | |
|---|---|---|---|---|
| CQ1 | 1.000 (1.000–1.000) | 0.872 (0.871–0.873) | 0.832 (0.831–0.834) | 0.844 (0.841–0.844) |
| CQ2 | 1.000 (1.000–1.000) | 0.816 (0.815–0.818) | 0.817 (0.815–0.818) | |
| CQ3 | 1.000 (1.000–1.000) | 0.992 (0.992–0.992) | ||
| CQ4 | 1.000 (1.000–1.000) |
Note: CQ1 = Inpatient Claims; CQ2 = Outpatient and Provider Claims with Procedure; CQ3 = Outpatient and Provider Claims with Fall; CQ4 = Inpatient or Outpatient and Provider Claims with Fall.
Discussion and Conclusions
Our study presents 4 different algorithms that can be used to identify FRIs using ICD-10 claims data. In general, the greatest number of events was identified using a broad approach that considers inpatient and outpatient events. The number of injuries that were added when outpatient claims were considered varies considerably by site: Hip fractures were generally identified using inpatient claims, whereas fractures of areas other than the hip and nonfracture injuries were more typically identified using outpatient and provider claims. Restricting hip fracture identification to the inpatient definition would decrease the number identified to 9–10th that identified by also incorporating outpatient claims. Additionally, not requiring external cause of injury codes allowed for identification of an additional 8% of hip fractures. Use of outpatient and provider claims definitions doubled the number of intracranial hemorrhages identified. While each algorithm captures varying numbers of FRIs, they do not appear to introduce systematic bias or uniquely underperform for any subpopulation as characteristics were similar between each of the 4 definitions. Expanding the diagnostic position to allow claims through the sixth position modestly increased the number of hip fractures detected in algorithms CQ1 and CQ3 but had little impact on algorithm CQ2.
Our study is novel in that it proposes multiple algorithms that are useful for detecting FRIs for residents presenting to different care settings after their presumed fall, recognizing that a single algorithm may be unsuitable for all purposes. For example, an investigator who is interested in studying FRIs resulting in hospitalizations may wish to employ the Inpatient algorithm, whereas an investigator seeking to capture any evidence of injury may wish to utilize the Inpatient and Outpatient and Provider with Procedure algorithms (CQ1 + CQ2). If an investigator was interested in increasing the likelihood that FRI claims treated medically or surgically were due to falls and not another mechanism of injury, the Outpatient and Provider with Fall algorithm (CQ3) might be appropriate. The CQ3 algorithm also allows for identification of FRIs where there is some uncertainty as to the precise diagnosis (eg, fall with hip pain and an outpatient diagnostic code for hip fracture without a confirmatory procedural code). If investigators want assurance that the injury occurred in the setting of a fall and not through another mechanism, the Inpatient or Outpatient and Provider with Fall algorithm (CQ4) exists. However, given that 8% of severe hospitalized hip fractures lacked a concomitant fall indicator, it may be preferable to use a combination of the Inpatient and Outpatient and Provider with Fall algorithms (CQ1 + CQ3) to capture all injuries with a suspected fall. By creating multiple algorithms, we hope that investigators, policymakers, and other stakeholders who use claims data to identify FRIs will be able to choose an algorithm that best matches their research question.
We are unaware of any prior studies that have developed multiple algorithms to define an FRI using ICD-10 claims data. Nevertheless, our work extends a similar study conducted in the Health and Retirement Study. In a large sample (n = 13 939), investigators developed 3 algorithms to identify FRIs using ICD-9 claims and then compared these algorithms with falls documented through E-codes or self-report (15). Investigators demonstrated that sensitivity of the algorithms increased as outpatient claims and less severe injuries (eg, sprains) were added, but at the expense of reduced PPV. Additionally, investigators found that FRIs identified using inpatient or emergency department claims seldom include an external cause of injury code for falls (58.0%), as compared with 92.5% of inpatient hip fractures in our study. The discrepancy in external cause of injury codes for falls we found compared to previous work may be due to the source of our data. Our study utilized linked MedPAR, Outpatient RIF, and Carrier RIF data, allowing identification of external cause of injury codes on the same day as FRI claims for many inpatient hip fractures. It is also important to note that, unlike the study by Min et al. (15), we did not use an episodic approach to FRI identification. A separate study comparing the detection of major injury falls by MDS with a claims-based algorithm using MedPAR showed external cause of injury code rates ranging from 86% to 94%, similar to our findings for hip fracture (33). Importantly, there are regional differences in reporting external cause of injury codes. The 2001 Healthcare Cost and Utilization Project found the highest compliance with reporting of these codes was in 8 of the 22 states with provisions for enforcing coding mandates (26,34,35). Investigators should be aware of these potential discrepancies if studying regional differences in FRI, as state-specific mandates and their enforceability may affect the algorithms that require external cause of injury codes. It is also important to note that inclusion of outpatient codes increases the rate of non-FRIs within the data set. It is notable that a large number of FRIs identified by CQ1 and CQ2 had external cause of injury codes associated with them, suggesting either the major nature of these injuries makes these codes unnecessary, or the efforts taken to increase external cause of injury code usage has been effective, consistent with recent Healthcare Cost and Utilization Project reporting (36).
Our FRI detection algorithms are designed to use inpatient claims, outpatient claims, or a combination of both, which gives the flexibility to detect injuries regardless of the treatment setting. Furthermore, we used an expansive set of ICD-10 codes to detect a wide array of FRIs including dislocations, vertebral column fractures, and closed head injuries, as one or all of such claim types been previously left out of other published algorithms (11–16). Our study analyzes a large representative sample of all the claims submitted to Medicare in 2016, removing selection bias that may occur when examining single or multiple facility data.
Limitations
This work must be viewed considering several potential limitations. First, we do not know the performance characteristics of our algorithms, including sensitivity and PPV, although this is highly important. Prior research has demonstrated that the MDS falls assessment has a low PPV compared with medical records (37,38). These studies demonstrate instances of overreporting from carrying forward a history of falls across the MDS assessments and underreporting of incident falls. Recent attempts to assess the MDS against validated ICD-9 algorithms have also shown poor performance with a correlation of 0.22 between the MDS assessment and claims-based algorithms (33). A study by Wright et al. (14) employed a multiple algorithm approach to identify 9 different fracture types using ICD-9 claims data. The investigators used 4 different CQ algorithms: inpatient diagnostic claim in the primary diagnosis, inpatient diagnostic claim in a secondary diagnosis, noninpatient diagnostic claim with procedural claim, or noninpatient claim with imaging. Fractures were validated in 520 participants. Overall, the PPV was good for most algorithms. For severe injuries like hip fracture, the PPV was higher using the inpatient algorithm rather than the noninpatient algorithm (96.7% vs 87.5%). For less severe injuries (eg, clavicle fracture), the PPV was lower using the inpatient versus noninpatient algorithm (66.7% vs 100.0%). We are unable to obtain electronic medical records data to validate our algorithms across health care settings, but we suspect that similar trends may exist.
Our study also relies upon the Residential History File—an algorithm combining claims and MDS assessment—to define the cohort and determine which residents qualify as long-stay (23). Other methods, such as that described by Yun et al. (39), utilized filed claims to make a decision about long-stay status, although some discrepancies exist when comparing these methods to the Residential History File. The proposed algorithms have not been validated in long-stay or other populations, and future validation studies should carefully specify the method used to identify long-stay residents.
Second, our algorithms rely heavily on accurate coding and thus are subject to the influence of widespread miscoding (ie, coding a fracture with a fall as a traumatic hip fracture or vice versa) in Medicare claims. This is especially important if making comparisons between disparate geographic regions, where discrepancies in coding could lead to different FRI rates. In this study, investigators captured only the first injury occurrence; however, some research questions may dictate measuring total injuries. Our present algorithm seeks to differentiate incident injuries from complications by requiring the use of an ending letter specifier signifying an initial claims episode and by limiting detection of FRIs to the primary diagnostic or first secondary claims position. Other investigators have proposed using care episodes as part of the algorithm to differentiate incident injuries, and it is likely prudent to use care episodes when validating these algorithms and when considering total injuries (14,15). Readers should be aware that some algorithms, particularly those relying solely on outpatient claims, may be less reliable. However, our algorithm provides an understanding of the relative frequency of how and where these injuries are treated.
Third, the present data included only major injuries, and minor injuries such as abrasions and lacerations were not included. Persons interested in describing the totality of FRI in NH residents may wish to add these codes, included in Supplementary Table 4.
Conclusions
We present a series of new algorithms that may be used to identify FRIs in Medicare claims data in NH residents. These algorithms overlap and investigators may elect to use different algorithms depending on the research questions asked. When studying the prevalence of major FRIs or identifying risk factors for FRI, we suggest using a combination of the Inpatient and Outpatient and Provider with Fall algorithms (CQ1 + CQ3), given the number of major FRIs that would be missed if an external cause of injury code were required on inpatient claims. Alternatively, the Inpatient and Outpatient Provider with Procedure (CQ1 + CQ2) definitions may be appropriate if investigators are more concerned with identifying specific injury types that could occur outside the setting of a fall. Despite the novelty of our designed algorithm for NH residents, caution should be utilized by investigators and policymakers as the algorithm is not currently validated in a representative sample of NH residents. We hope the FRI algorithms described here will be an important tool for future work among NH residents. We encourage investigators to clearly specify the algorithm used to identify FRIs in administrative claims data going forward.
Supplementary Material
Contributor Information
Joel Mintz, Nova Southeastern University College of Allopathic Medicine, Davie, Florida, USA; Hinda and Arthur Marcus Institute for Aging Research, Hebrew SeniorLife, Roslindale, Massachusetts, USA.
Matthew S Duprey, Department of Health Services, Policy, and Practice, Brown University, Providence, Rhode Island, USA.
Andrew R Zullo, Department of Health Services, Policy, and Practice, Brown University, Providence, Rhode Island, USA; Center of Innovation in Long-Term Services and Supports, Providence Veterans Affairs Medical Center, Providence, Rhode Island, USA.
Yoojin Lee, Department of Health Services, Policy, and Practice, Brown University, Providence, Rhode Island, USA.
Douglas P Kiel, Hinda and Arthur Marcus Institute for Aging Research, Hebrew SeniorLife, Roslindale, Massachusetts, USA; Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts, USA.
Lori A Daiello, Department of Health Services, Policy, and Practice, Brown University, Providence, Rhode Island, USA.
Kenneth E Rodriguez, Department of Orthopedic Trauma Surgery, Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA.
Arjun K Venkatesh, Department of Emergency Medicine, Yale University School of Medicine, New Haven, Connecticut, USA.
Sarah D Berry, Hinda and Arthur Marcus Institute for Aging Research, Hebrew SeniorLife, Roslindale, Massachusetts, USA; Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts, USA.
Funding
This work was supported by the National Institute on Aging (grant numbers RF1AG061221, R21AG061632, and R01AG065722 to A.R.Z.) and the National Institute of General Medical Sciences (grant number U54GM115677 to A.R.Z.) and Medical Student Training in Aging Research Grant 5T35AG038027-09. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflict of Interest
None declared.
Author Contributions
J.M., A.R.Z., Y.L., D.P.K., L.A.D., K.E.R., A.K.V., and S.D.B. were responsible for the design of the study. J.M., A.R.Z., and S.D.B. were responsible for data acquisition. J.M., M.S.D., A.R.Z., Y.L., and S.D.B. were responsible for data analysis. All authors participated in data interpretation, drafting and revising the manuscript, and approving the final version.
References
- 1. Rubenstein LZ, Josephson KR, Robbins AS. Falls in the nursing home. Ann Intern Med. 1994;121(6):442–451. doi: 10.7326/0003-4819-121-6-199409150-00009 [DOI] [PubMed] [Google Scholar]
- 2. Florence CS, Bergen G, Atherly A, Burns E, Stevens J, Drake C. Medical costs of fatal and nonfatal falls in older adults. J Am Geriatr Soc. 2018;66(4):693–698. doi: 10.1111/jgs.15304 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Berry SD, Lee Y, Zullo AR, Kiel DP, Dosa D, Mor V. Incidence of hip fracture in U.S. nursing homes. J Gerontol A Biol Sci Med Sci. 2016;71(9):1230–1234. doi: 10.1093/gerona/glw034 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Rapp K, Becker C, Lamb SE, Icks A, Klenk J. Hip fractures in institutionalized elderly people: incidence rates and excess mortality. J Bone Miner Res. 2008;23(11):1825–1831. doi: 10.1359/jbmr.080702 [DOI] [PubMed] [Google Scholar]
- 5. Ouslander JG, Lamb G, Perloe M, et al. Potentially avoidable hospitalizations of nursing home residents: frequency, causes, and costs: [see editorial comments by Drs. Jean F. Wyman and William R. Hazzard, pp 760–761]. J Am Geriatr Soc. 2010;58(4):627–635. doi: 10.1111/j.1532-5415.2010.02768.x [DOI] [PubMed] [Google Scholar]
- 6. Tinetti ME, Williams CS. Falls, injuries due to falls, and the risk of admission to a nursing home. N Engl J Med. 1997;337(18):1279–1284. doi: 10.1056/NEJM199710303371806 [DOI] [PubMed] [Google Scholar]
- 7. Neuman MD, Silber JH, Magaziner JS, Passarella MA, Mehta S, Werner RM. Survival and functional outcomes after hip fracture among nursing home residents. JAMA Intern Med. 2014;174(8):1273–1280. doi: 10.1001/jamainternmed.2014.2362 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Drew JAR, Xu D. Trends in fatal and nonfatal injuries among older Americans, 2004–2017. Am J Prev Med. 2020;59(1):3–11. doi: 10.1016/j.amepre.2020.01.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Keene GS, Parker MJ, Pryor GA. Mortality and morbidity after hip fractures. BMJ. 1993;307(6914):1248–1250. doi: 10.1136/bmj.307.6914.1248 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Schnell S, Friedman SM, Mendelson DA, Bingham KW, Kates SL. The 1-year mortality of patients treated in a hip fracture program for elders. Geriatr Orthop Surg Rehabil. 2010;1(1):6–14. doi: 10.1177/2151458510378105 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Kim SB, Zingmond DS, Keeler EB, et al. Development of an algorithm to identify fall-related injuries and costs in Medicare data. Inj Epidemiol. 2016;3(1):1. doi: 10.1186/s40621-015-0066-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Ray WA, Griffin MR, Fought RL, Adams ML. Identification of fractures from computerized Medicare files. J Clin Epidemiol. 1992;45(7):703–714. doi: 10.1016/0895-4356(92)90047-q [DOI] [PubMed] [Google Scholar]
- 13. Waters TM, Chandler AM, Mion LC, et al. Use of international classification of diseases, ninth revision, clinical modification, codes to identify inpatient fall-related injuries. J Am Geriatr Soc. 2013;61(12):2186–2191. doi: 10.1111/jgs.12539 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Wright NC, Daigle SG, Melton ME, Delzell ES, Balasubramanian A, Curtis JR. The design and validation of a new algorithm to identify incident fractures in administrative claims data. J Bone Miner Res. 2019;34(10):1798–1807. doi: 10.1002/jbmr.3807 [DOI] [PubMed] [Google Scholar]
- 15. Min L, Tinetti M, Langa KM, Ha J, Alexander N, Hoffman GJ. Measurement of fall injury with health care system data and assessment of inclusiveness and validity of measurement models. JAMA Netw Open. 2019;2(8):e199679. doi: 10.1001/jamanetworkopen.2019.9679 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Hoffman GJ, Hays RD, Shapiro MF, Wallace SP, Ettner SL. Claims-based identification methods and the cost of fall-related injuries among US older adults. Med Care. 2016;54(7):664–671. doi: 10.1097/MLR.0000000000000531 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Office of the Secretary, HHS. Administrative simplification: change to the compliance date for the International Classification of Diseases, 10th Revision (ICD-10-CM and ICD-10-PCS) medical data code sets. Final rule. Fed Regist. 2014;79(149):45128–45134. [PubMed] [Google Scholar]
- 18. Neuman MD, Fleisher LA, Even-Shoshan O, Mi L, Silber JH. Nonoperative care for hip fracture in the elderly: the influence of race, income, and comorbidities. Med Care. 2010;48(4):314–320. doi: 10.1097/mlr.0b013e3181ca4126 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Berry SD, Zullo AR, McConeghy K, Lee Y, Daiello L, Kiel DP. Administrative health data: guilty until proven innocent. Response to comments by Levy and Sobolev. Osteoporos Int. 2018;29(1):255–256. doi: 10.1007/s00198-017-4244-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Morris JN, Hawes C, Fries BE, et al. Designing the national resident assessment instrument for nursing homes. Gerontologist. 1990;30(3):293–307. doi: 10.1093/geront/30.3.293 [DOI] [PubMed] [Google Scholar]
- 21. Thomas KS, Dosa D, Wysocki A, Mor V. The Minimum Data Set 3.0 cognitive function scale. Med Care. 2017;55(9):e68–e72. doi: 10.1097/MLR.0000000000000334 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Saliba D, Jones M, Streim J, Ouslander J, Berlowitz D, Buchanan J. Overview of significant changes in the Minimum Data Set for nursing homes version 3.0. J Am Med Dir Assoc. 2012;13(7):595–601. doi: 10.1016/j.jamda.2012.06.001 [DOI] [PubMed] [Google Scholar]
- 23. Intrator O, Hiris J, Berg K, Miller SC, Mor V. The residential history file: studying nursing home residents’ long-term care histories(*). Health Serv Res. 2011;46(1 Pt 1):120–137. doi: 10.1111/j.1475-6773.2010.01194.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Berry SD, Zullo AR, McConeghy K, Lee Y, Daiello L, Kiel DP. Defining hip fracture with claims data: outpatient and provider claims matter. Osteoporos Int. 2017;28(7):2233–2237. doi: 10.1007/s00198-017-4008-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Casey MM, Moscovice IS, Virnig BA, Durham SB. Providing hospice care in rural areas: challenges and strategies. Am J Hosp Palliat Care. 2005;22(5):363–368. doi: 10.1177/104990910502200509 [DOI] [PubMed] [Google Scholar]
- 26. Annest JL, Fingerhut LA, Gallagher SS, et al. Strategies to improve external cause-of-injury coding in state-based hospital discharge and emergency department data systems: recommendations of the CDC Workgroup for Improvement of External Cause-of-Injury Coding. MMWR Recomm Rep. 2008;57(RR-1):1–15. https://www.cdc.gov/mmwr/preview/mmwrhtml/rr5701a1.htm [PubMed] [Google Scholar]
- 27. Morris JN, Fries BE, Morris SA. Scaling ADLs within the MDS. J Gerontol A Biol Sci Med Sci. 1999;54(11):M546–M553. doi: 10.1093/gerona/54.11.m546 [DOI] [PubMed] [Google Scholar]
- 28. Carpenter GI, Hastie CL, Morris JN, Fries BE, Ankri J. Measuring change in activities of daily living in nursing home residents with moderate to severe cognitive impairment. BMC Geriatr. 2006;6:7. doi: 10.1186/1471-2318-6-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Guilford JP. The phi coefficient and chi square as indices of item validity. Psychometrika. 1941;6(1):11–19. doi: 10.1007/BF02288569 [DOI] [Google Scholar]
- 30. Davenport EC, El-Sanhurry NA. Phi/Phimax: review and synthesis. Educ Psychol Meas. 1991;51(4):821–828. doi: 10.1177/001316449105100403 [DOI] [Google Scholar]
- 31. Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas. 1960;20(1):37–46. doi: 10.1177/001316446002000104 [DOI] [Google Scholar]
- 32. McHugh ML. Interrater reliability: the kappa statistic. Biochem Med (Zagreb). 2012;22(3):276–282. doi: 10.11613/BM.2012.031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Sanghavi P, Pan S, Caudry D. Assessment of nursing home reporting of major injury falls for quality measurement on nursing home compare. Health Serv Res. 2020;55(2):201–210. doi: 10.1111/1475-6773.13247 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Barrett M, Steiner C, Coben J.. Healthcare Cost and Utilization Project (HCUP) E Code Evaluation Report. 2004. US Agency for Healthcare Research and Quality; 2005. doi: 10.4135/9781483349985.n182 [DOI] [Google Scholar]
- 35. Coben JH, Steiner CA, Barrett M, Merrill CT, Adamson D. Completeness of cause of injury coding in healthcare administrative databases in the United States, 2001. Inj Prev. 2006;12(3):199–201. doi: 10.1136/ip.2005.010512 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Healthcare Cost and Utilization Project. Injuries and External Causes: Reporting of Causes on the HCUP State Inpatient Databases, 2016–2017. Agency for Healthcare Research and Quality; 2020. www.hcupus.ahrq.gov/reports/ataglance/findingsataglance.jsp [PubMed] [Google Scholar]
- 37. Survey and Certification Group, Centers for Medicare and Medicaid Services. MDS/staffing focused surveys update. 2015. https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/SurveyCertificationGenInfo/Downloads/Survey-and-Cert-Letter-15-25.pdf. Accessed March 18, 2021.
- 38. Mintz J, Lee A, Gold M, Hecker EJ, Colón-Emeric C, Berry SD. Validation of the minimum data set items on falls and injury in two long-stay facilities. J Am Geriatr Soc. 2021;69(4):1099–1100. doi: 10.1111/jgs.16974 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Yun H, Kilgore ML, Curtis JR, et al. Identifying types of nursing facility stays using Medicare claims data: an algorithm and validation. Health Serv Outcomes Res Method. 2010;10(1–2):100–110. doi: 10.1007/s10742-010-0060-4 [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.


