The parent Health ABC study enrolled 3,075 women and men, aged 70–79, from two field centers, Pittsburgh, PA and Memphis, TN at visit 1 (baseline, 1997–1998). Participants had to report no difficulty walking at least 1/4 mile and or climbing a flight of stairs to be eligible to participate. White participants were identified from a random sample of white Medicare beneficiaries. Black subjects were identified as age-eligible community residents from designated ZIP code areas surrounding Pittsburgh PA and Memphis TN. Exclusion criteria included reported difficulty performing basic activities of daily living, obvious cognitive impairment, inability to communicate with the interviewer, intention of moving within 3 years, or participation in a trial involving a lifestyle intervention. There were 3,044 enrollees that remained at visit 2 (1998–1999).

The Health ABC Knee Osteoarthritis Substudy included 1123 participants from visits 2 (1998–1999) or 3 (1999–2000) (Figure). Participants were included in the knee Osteoarthritis Substudy at visit 2 if they had qualifying knee pain and knee x-ray. Cases with qualifying knee pain were identified if they had “knee pain, aching, or stiffness on most days for at least 1 month” at some point over the previous year or if they reported moderate or worse knee pain during the previous month in association with at least one activity on the Western Ontario and McMaster Universities Arthritis Index (WOMAC) knee pain scale (34). Additional participants with qualifying pain were added at visit 3 if they had qualifying knee pain at both visits 2 and 3, and knee x-ray at visit 3. Finally, a random set of 270 controls (with no qualifying pain at either visit 2 or 3) with knee x-ray at visit 3 were selected from 1798 participants for inclusion in the knee Osteoarthritis Substudy (35).

Participants included in the analytic sample (n=734) were those that were followed after study baseline (i.e., visits 2 or 3 in 1998–2000), with a clinic visit 4 (Medicare data was only collected for those with a clinic visit 4 (2000–2001)), did not have a missing K/L grade reading of their knee x-ray, and had Medicare fee-for-service (FFS) in CMS (to ascertain injurious falls status) that extended beyond study baseline enrollment date (1998–2000) (Figure). Participants in the analytic study were followed from study baseline (1998–2000) until the occurrence of an injurious fall, the loss of FFS, loss to follow-up, death, or through visit 11 (2007–2008) (when Medicare claims were last collected for this study). Injurious fall status was assessed during this time period of FFS in the CMS dataset. The institutional review board at each center approved the study protocol, and written informed consent was obtained from all the participants.

At both visits 2 and 3, expert readers assessed posterior-anterior and skyline projection knee x-rays to assess Kellgren-Lawrence (K/L) grade based on individual radiographic features (joint space narrowing, osteophytes, subchondral attrition, cysts and sclerosis) and scored using the Osteoarthritis Research Society International atlas in the medial and lateral compartment of the tibio-femoral joint and the patello-femoral joint (36). Participants had x-rays at either visits 2 or 3. Follow-up began at either V2 or V3 depending on when the x-ray was taken. Inter-rater reliability was excellent (weighted kappa 0.87 for K/L grade). Knee ROA was defined as a K/L grade ≥ 2.

Adults were categorized into 4 mutually exclusive groups: knee sROA, knee ROA without pain, knee pain without ROA, and no ROA or pain. During the clinic visit when knee imaging was completed, participants were asked if they had, “knee pain on most days in the past 30 days”. Knee pain was defined as having pain symptoms (during the majority of the last 30 days) in at least one knee. Knee ROA was defined as having a Kellgren-Lawrence grade of ≥2 in at least one knee. Knee sROA was defined as having both ROA and pain symptoms in the same knee.

Incident injurious falls were ascertained from outpatient and inpatient Medicare claims and defined using a diagnosis code algorithm from linked Medicare claims. All injuries captured in Medicare claims are included, which includes any outpatient care billed by any type of provider. Any unique event with an ICD-9 fall code (E880–888) plus non-fracture injury, a vertebral fracture code (805–806) with a fall code, or any non-vertebral fracture code (800–804, 807–829) with/without a fall code was considered an injurious fall as an estimated 80% of non-vertebral fractures are attributed to falls (37). All traumatic injuries (e.g. motor vehicle accidents), intentional injuries and pathologic injuries were not considered to meet the definition of an injurious fall.

In the adjudication of our diagnoses code algorithm as previously detailed (38), a subset of Medicare fall injuries were compared to the self-reported fall injuries with medical records. The injuries adjudicated were included due to potential uncertainty of particular diagnoses codes to classify a primary injurious fall: concurrent stroke code; fall code with uncertain injury; fracture code that was not in 1^st or 2^nd position (i.e., not listed as the 1^st or 2^nd reason for visit/hospitalization billing); vertebral column/rib fracture without a fall code. Overall, the injurious fall adjudication showed an excellent agreement, except for vertebral fractures where only 50% of vertebral fractures were confirmed. We revised the initial diagnoses code algorithm to exclude vertebral fractures without concurrent fall codes.

All covariates evaluated for inclusion in models were measured once at either visit 1 or 2. Potential covariates associated with the exposure or outcome at p<0.1 were included in the full multivariate adjusted model. If covariates were available at both visits the visit one measurement was used. These potential covariates were selected based on documented associations with knee OA and falls. Demographic variables included self-report of age, sex, race (white or black), and education (<high school (HS), HS graduate, or postsecondary), and study site (Memphis or Pittsburgh). Weight was measured on a standard balance beam scale to the nearest 0.1 kg. Height was measured by a stadiometer to the nearest 0.1 cm. The anthropometric measure of BMI (kg/m²) was calculated using the formula weight (kg)/height² (m²).

Lifestyle factors included self-report of smoking (never, past smoker, current smoker) and physical activity (kcal/kg/wk). Physical activity was determined using the caloric expenditure in the past week for self-reported duration of walking, climbing stairs, and exercise (39).

Several medical characteristics were considered for the analysis. Participants self-reported their current health status (fair/poor/very poor versus good/very good/excellent) and history of falls in the past 12 months, depression, poor vision, myocardial infarction, and stroke. Diabetes was defined using fasting glucose (≥126 mg/dl), self-report, or hypoglycemic medication use. Diagnosed and/or treated hypertension (hypertension hereafter) was defined via self-report or antihypertensive medication use. To assess supplementary intake for vitamin D and calcium, and medication use such as non-steroidal anti-inflammatory drugs (NSAIDs), statins, steroid use, and antidepressants, participants were asked to bring all prescription and over the counter medications, which were coded based on the Iowa Drug Information System (40). Total number of other medications was assessed using number of other prescription medications (excluding steroids & antidepressants).

Chi-square tests were used to evaluate proportion differences for incident injurious falls across study covariates. Two-sample t-tests (Wilcoxon rank-sum test for non-normally distributed data) were used to examine mean differences in continuous covariates by incident injurious falls status. To compare baseline knee OA status by study covariates chi-square tests were performed to assess proportion differences. Fisher’s exact test was performed for all tests of proportion if the expected value for any cell was <5. To compare mean differences in continuous covariates by baseline knee OA status ANOVA (Kruskal-Wallis test for non-normally distributed data) was performed.

Cox proportional-hazards models were used to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) and compare the time from the visits 2 or 3 (depending on when x-ray was taken) to incident injurious falls by baseline knee OA status (knee sROA, knee pain without ROA, knee ROA without pain groups compared with reference group (no ROA or pain)), while controlling for potential confounders. Participants were right censored if they did not have the event of interest by the time they were lost to follow-up, their follow-up ended, or by the time of their death. Individuals with missing covariate data were dropped from the multivariate analyses. Furthermore, we performed a priori secondary analyses stratified by sex and by injurious falls type (fracture versus non-fracture). The proportional hazards assumption was assessed by testing (1) the interaction of knee OA indicator variables (sROA, ROA without pain, or pain without ROA, vs. no ROA/no pain) with log(time) and (2) the knee OA indicator variables using the Supremum test based on Schoenfeld residuals, in full multivariate adjusted models. The proportional hazards assumption was considered violated if p<0.05 for the knee OA indicator variables*log(time) interaction terms or knee OA indicator variables. Knee OA did not violate the proportional hazards assumption in the full multivariate adjusted overall, sex-stratified and injury type-stratified models.

All analyses were performed using the Statistical Analysis System (SAS, version 9.3; SAS Institute, Cary, NC, USA). Statistical significance for all analyses was determined at the α<0.05 level.

The mean (SD) follow-up time was 6.59 (3.12) years. The mean (SD) age of the participants was 74.7 (2.9) years. There were 249 participants with sROA, 32 had ROA without pain, 306 had pain without ROA, and 147 without pain or ROA in either knee. Of the 734 participants, 255 (34.7%) had an incident injurious fall over the entire study period. The average annual incidence rate of injurious falls across 11 years (from 1998 through 2008) was 4.84 per 100 person-years (p-yrs). The incidence rate per 100 p-yrs of injurious falls by year was 0.7 (1998), 3.0 (1999), 4.0 (2000), 5.8 (2001), 5.2 (2002), 3.7 (2003), 4.8 (2004), 5.4 (2005), 3.2 (2006), 6.8 (2007), and 10.3 (2008). Table 1 shows the distributions of baseline knee OA status and incident injurious falls by baseline characteristics. Baseline knee OA status differed significantly by race, site, health status, history of falls, hypertension status, NSAID use, or steroid use (Table 1). Women, whites, those who had higher education, never smoked, better health, or no heart disease, or those who took calcium or vitamin D supplements, anti-depressants, or steroids were significantly more likely to have an incident injurious fall by end of follow-up (Table 1). Mean age and BMI varied significantly by knee OA status (p<0.01 for both). Baseline mean age was higher (75.1 vs. 74.5 years, p=0.01) and baseline mean BMI was lower (27.3 vs. 28.3 kg/m², p=0.01) among adults with an incident injurious fall by end of follow-up. Median physical activity did not significantly differ by knee ROA status (p=0.35). Baseline median physical activity was 2.4 kcal/kg/week for those with an incident injurious fall and 2.2 kcal/kg/week for those without an incident injurious by end of follow-up, p=0.09. Number of other prescription medications (excluding steroids & antidepressants) differed by knee OA status (knee sROA=3.64, knee ROA without pain=2.81, pain with knee ROA=3.74, no knee pain or ROA=2.51, p<0.01) and was borderline significantly higher among adults with an injurious fall (3.61 vs. 3.18, p=0.05).

There were a total of 77 injurious falls out of 288 (26.7%) men, and the cumulative incidence varied by knee OA status. The cumulative incidence of injurious falls by knee OA group was 33.3% for sROA, 16.7% for ROA without pain, 29.4% for pain without ROA, and 14.9% for no ROA or pain.

There were 178 injurious falls out of 446 (39.9%) participants, and the cumulative incidence did not vary by knee OA status. The cumulative incidence of injurious falls by knee OA group was 36.5% for sROA, 35.0% for ROA without pain, 43.9% for pain without ROA, and 38.8% for no ROA or pain.

In the multivariate model with men and women combined, compared with those without ROA or pain, individuals with sROA (HR=1.09; 95% CI: 0.73, 1.65), ROA without pain (HR=1.01; 95% CI: 0.46, 2.20), and pain without ROA (HR=1.08; 95% CI: 0.74, 1.57) did not have a significantly increased risk of injurious falls (Table 2). Among men only, and compared with men without ROA or pain, those with sROA (HR=2.57; 95% CI: 1.12, 5.91) had a significantly higher risk of injurious falls (Table 2). No significant association existed between knee OA and injurious falls in women. The 4*2 interaction term predicting injurious falls between knee OA and sex was not statistically significant (p>0.05); however, interpretation of this result should be viewed with caution as this analysis was likely underpowered as a result of the low number of adults with ROA and no pain. The association between knee OA and injurious falls did not differ by injurious fall type (fracture versus non-fracture) (Table 3).