Associations of Days Spent at Home Before Hip Fracture With Postfracture Days at Home and 1-Year Mortality Among Medicare Beneficiaries Living With Alzheimer’s Disease or Related Dementias

Abstract

Background

Hip fracture is a disabling event experienced disproportionately by older adults with Alzheimer’s disease or related dementias (ADRD). Claims information recorded prior to a hip fracture could provide valuable insights into recovery potential for these patients. Thus, our objective was to identify distinct trajectories of claims-based days at home (DAH) before a hip fracture among older adults with ADRD and evaluate associations with postfracture DAH and 1-year mortality.

Methods

We conducted a cohort study of 16 576 Medicare beneficiaries living with ADRD who experienced hip fracture between 2010 and 2017. Growth mixture modeling was used to estimate trajectories of DAH assessed from 180 days prior to fracture until index fracture admission, and their joint associations with postfracture DAH trajectories and 1-year mortality.

Results

Before a hip fracture, a model with 3 distinct latent DAH trajectories was the best fit. Trajectories were characterized based on their temporal patterns as Consistently High (n = 14 980, 90.3%), Low but Increasing (n = 809, 5.3%), or Low and Decreasing (n = 787, 4.7%). Membership in the Low and Decreasing prefracture DAH trajectory was associated with less favorable postfracture DAH trajectories, and a 65% higher 1-year mortality rate (hazard ratio 1.65, 95% confidence interval 1.45–1.87) as compared to those in the Consistently High trajectory. Similar albeit weaker associations with these outcomes were observed for hip fracture survivors in the Low but Improving prefracture DAH trajectory.

Conclusions

Distinct prefracture DAH trajectories among hip fracture survivors with ADRD are strongly linked to postfracture DAH and 1-year mortality, which could guide development of tailored interventions.

Nearly 300 000 older adults experience a hip fracture every year, the majority of which occur after a fall (1). Hip fractures are highly disabling, and recovery is often incomplete (2). Fall-related fractures are associated with a threefold increase in the risk of long-stay nursing home admission (3), and hip fracture survivors spend 47 more days homebound or receiving care in a health care facility than their healthy peers annually (4). Older adults living with Alzheimer’s disease and related dementia (ADRD) experience worse outcomes after hip fracture (2,5,6), which is concerning given the rapidly increasing population of older adults living with ADRD in the United States. But despite common misconceptions (7), many older adults living with ADRD have the potential to improve with rehabilitation after hip fracture, and some may even experience robust improvements when given the opportunity to receive and participate in intensive interventions (8,9). However, rehabilitation providers often struggle to identify which patients with ADRD are most likely to benefit (7).

Prior work has highlighted the value of prefracture functional information in prognosticating postfracture outcomes (2,10), including the importance of prefracture status in identifying those most likely to experience functional recovery (11). However, these studies have not focused on older adults with ADRD; better understanding relationships between prefracture and postfracture function among older adults with ADRD could guide more tailored discharge planning (eg, rehabilitation hospital vs nursing home), home and community-based rehabilitation programs, or referrals for additional services such as palliative care.

However, prefracture data on function are rarely available in medical records and may be difficult to obtain during urgent admissions for a traumatic injury. Surrogate measures of prefracture function beyond comorbidities, thus, may provide valuable information. One such measure is a claims-based measure of days at home (DAH). Spending fewer DAH is associated with poorer self-rated health, mobility and self-care difficulties, and limited social activity among older adult populations (12). While DAH has been used to measure quantity and quality of postfracture survival (13), no studies have examined heterogeneity of prefracture DAH information or the associations with postfracture recovery. DAH data such as hospital days and nursing home days spent in the prior year are readily available to hospitals as part of Medicare eligibility verifications and thus could be a valuable additional resource to guide treatment decision making and discharge planning. This may especially be of benefit to older adults with ADRD who may not be able to provide accurate prefracture information during a hospitalization for a traumatic injury.

Therefore, the objectives of this study were to identify distinct prefracture DAH trajectories among older adults living with ADRD, and evaluate associations of these trajectories with hip fracture recovery. We hypothesized that membership in unfavorable prefracture DAH trajectories would be strongly associated with poorer postfracture DAH trajectories and higher 1-year mortality.

Method

Data Source

Data were drawn from a 5% random sample of claims and assessment data for Medicare beneficiaries (2010–2016) and a 20% random sample of Medicare beneficiaries (2017–2018). The Medicare Provider and Analysis Review (MEDPAR) file was used to identify hip fracture index admissions that occurred between July 1, 2010 and December 31, 2017. We additionally used data from the Medicare Master Beneficiary Summary File (MBSF), Minimum Data Set (MDS) 3.0, Chronic Conditions, Outpatient claims, and Carrier claims to capture demographic, comorbidity, and health care utilization data for the cohort. This study of Medicare claims data was deemed exempt, non-human subjects research by the University of Maryland Baltimore Institutional Review Board.

Study Sample

Patients living with a diagnosis of ADRD and continuous Medicare fee-for-service enrollment 6 months prior to and 12 months following an index hip fracture admission date were included in our study cohort. An ADRD diagnosis was identified in Medicare claims using the Chronic Conditions file (14). Index hip fractures were defined as an inpatient hospital admission with an admission International Classification of Diseases, 9th revision (ICD-9) or International Classification of Diseases, 10th edition (ICD-10) diagnosis code corresponding to a femoral neck or intertrochanteric fracture of the hip (Supplementary Methods 1) (15). If a patient had multiple qualifying hip fractures, the first was considered the index hospitalization. We excluded long-stay nursing home residents prior to admission (16), those who died during admission or were discharged against medical advice, and those who were hospitalized in a U.S. territory (eg, Puerto Rico).

Identifying DAH in Medicare Claims

The DAH measure was based on prior work in surgical populations that assessed days alive and out of the hospital (17–19) and was further validated in other medically complex older adult populations (12). Briefly, for each patient-month of the study period, home time was calculated by subtracting the total number of days in hospitals, skilled nursing facilities, days in hospital observation, and emergency department days from the total number of days alive. MEDPAR data were used to identify hospitalizations (including short-stay acute care hospitalizations, long-term acute care hospitalizations, and inpatient rehabilitation stays). Both short-stay and long-stay nursing home days were identified using MDS data. Days in the emergency department or under hospital observation, consistent with prior work, were extracted from Medicare Outpatient claims files using revenue codes associated with each setting (4). From 6 months prior to hip fracture to 12 months following the index fracture admission, we calculated the count of DAH spent each month. If a patient died during any follow-up month, the total number of DAH was considered missing for subsequent months until the end of the follow-up period.

Variables

Demographic variables including patient age, sex, self-reported racial and ethnic identity, county of residence, and Medicaid dual eligibility were extracted from the Medicare MBSF. Hospital length of stay, including whether the patient stayed in the intensive care unit (ICU), was extracted from the MEDPAR file using admission and discharge dates, and indicator codes for ICU claims. Surgical procedure codes were used to categorize the hip fracture repair as an arthroplasty (hemiarthroplasty or total hip replacement), open reduction with internal fixation, or nonsurgical management (including closed reductions) consistent with prior work (15). Multimorbidity was characterized using the Elixhauser index (range: 0–28, higher is more comorbidities), which was extracted from ICD-9 and ICD-10 diagnosis codes present in Medicare claims during the 6 months before hip fracture (20). We also captured characteristics that could have influenced DAH before or after fracture, including hospital-level factors (number of hospital beds, teaching status) and county-level (based on the patient’s zip code) poverty rates, nursing home beds per 100 000 residents, and rurality from rural–urban continuum codes provided by the U.S. Department of Agriculture.

Statistical Analysis

To identify trajectories of days spent at home before and after a hip fracture, we used joint trajectory analysis. This method estimates the likelihood of membership in each of the identified trajectories before and after a fracture and assigns participants to the trajectory with the highest probability of membership. A similar modeling strategy, with smaller cohort study data, has been used in other studies of recovery from fall-related trauma (11). We used Proc TRAJ in SAS (SAS Institute, Cary, NC) to estimate these trajectories, fitting a mixture model to our longitudinal data using a maximum likelihood method. For our models, DAH during each patient-month were assumed to follow a truncated normal distribution, with minimum and the maximum values set to be 0 and 30, respectively. The mean structures of DAH outcomes at specific time point at each month were modeled by cubic polynomial curves, as these models had better statistical fit compared to linear or quadratic curves. We first fit separate trajectory models to prefracture period (6 months before index fracture) and postfracture group (12 months following fracture) (21). To determine the optimal number of latent trajectories in each model (22), we used the change in the Bayesian Information Criterion (BIC) between models as an approximation to the log of the Bayes factor (23); lower BIC was indicative of better model fit. For prefracture and postfracture periods, we first fit a model with 1-class solution, and subsequently tested whether adding additional classes improved the model fit as assessed by BIC. For each model step, we weighed improved statistical fit against clinical interpretability consistent with established guidelines (24). To ensure our models included only clinically meaningful subgroups, we additionally restricted our consideration of additional class solutions to situations where membership within each class was estimated as ≥3% of the sample. Trajectory groups were named based on their temporal patterns. Posterior probabilities, or the accuracy of the proposed final models, were then calculated for each prefracture DAH trajectory and postfracture resilience trajectory to verify they appropriately fit the data. In general, posterior probabilities with an average value ≥0.9 within each trajectory are considered an excellent fit and values of <0.7 are considered a poor fit (23).

We then jointly modeled the number of DAH in the months before and after hip fracture. To do this, we first calculated unadjusted probabilities and 95% confidence intervals (CIs) of membership in postfracture trajectories conditional on membership in a given prefracture trajectory (21). CIs were approximated using a first-order Taylor series expansion. To adjust for participant, hospital, and geographic heterogeneity, we then used multinomial logistic regression to regress postfracture trajectories on prefracture trajectories and covariates (25). Covariates included age, sex, hospital length of stay and ICU use, Medicaid dual eligibility, Elixhauser score, surgical procedure, days since dementia diagnosis, hospital teaching status, year of fracture, and county-level measures of poverty, nursing home bed availability, and rurality based on the prefracture Medicare records for each survivor living with ADRD.

In addition, we assessed unadjusted associations of membership in prefracture trajectories with time-to-death over the year following fracture and estimated cumulative incidence curves for each prefracture DAH trajectory. Mortality hazard ratios comparing prefracture trajectory groups were estimated using an adjusted Cox proportional hazards regression with follow-up censored at 365 days (R Version 4.2.1, R Foundation for Statistical Computing, Vienna, Austria). This model was adjusted for the same covariates used in the multinomial regression model. The proportional hazards assumption was checked by statistical tests and graphical diagnostics tools based on the Schoenfeld residuals. In all statistical analyses, a p value ≤ .05 was considered to be statistically significant. Data analysis was conducted from December 2021 to April 2023.

Results

We identified 16 576 community-dwelling older adults living with ADRD who met the inclusion criteria. Over 60% of the sample was aged 85 or older, the majority identified as non-Hispanic White, and were female sex, and approximately 1 in 5 were dually eligible for Medicaid (Table 1). The mean (standard deviation [SD]) length of stay in the hospital was 5.5 (3.6) days, with 7% of the sample requiring ICU admission. Most hip fracture survivors with ADRD underwent open reduction and internal fixation of their fracture, and over 90% were discharged to nonhome location—predominantly skilled nursing facilities (70%) and inpatient rehabilitation facilities (14%).

Table 1.

Sample Demographics

Variable	Overall Sample (N = 16 576)
Age at hip fracture admission date, n (%)
65–74	1 129 (6.8)
75–84	5 338 (32.2)
85+	10 109 (61.0)
Age, years, mean (SD)	85.7 (6.9)
Female sex, n (%)	12 608 (76.1)
Race/ethnicity, n (%)
Non-Hispanic White	15 163 (91.5)
Black	712 (4.3)
Hispanic and Other*	671 (4.0)
Dual Medicaid–Medicare eligible, n (%)	3 490 (21.1)
Elixhauser comorbity index, mean (SD)	3.1 (1.8)
Time since dementia diagnosis, days, mean (SD)	1 940.4 (1 330.6)
ICU admission/stay, n (%)	1 159 (7.0)
Fracture management, n (%)
Partial or total arthroplasty	5 720 (34.5)
Internal fixation	9 292 (56.1)
Nonsurgical or other management	1 564 (9.4)
Hospital length of stays, days, mean (SD)	5.5 (3.6)
Discharge disposition, n (%)
Home or home health care	1 032 (6.2)
Facility†	15 544 (93.8)
For-profit hospital, n (%)	2 534 (15.3)
Teaching hospital, n (%)	7 768 (46.9)
Index hospital bed size, tertiles, n (%)
0–230 beds	5 468 (33.0)
231–424 beds	5 587 (33.7)
425–2 449 beds	5 519 (33.3)
Rurality, n (%)‡
Urban	13 580 (81.9)
Rural, urban adjacent	1 963 (11.8)
Rural, not urban adjacent	1 018 (6.1)
% of residents living below poverty line (tertiles), n (%)†
<12.5	5 741 (34.6)
12.5–17	5 544 (33.5)
>17	5 276 (31.8)
Number of primary care physicians per 100 000 (tertiles), n (%)†
<59	5 245 (31.6)
59–84	5 701 (34.4)
>84	5 567 (33.6)

Notes: ICU = intensive care unit; SD = standard deviation.

*Other races and ethnicities included those who were identified in Medicare claims data as Asian, Native Hawaiian and Pacific Islander, American Indian/Alaska Native, or those who identified as belonging to multiple racial subgroups.

^†Facility includes skilled nursing facilities, inpatient rehabilitation facilities, swing beds, long-term acute care hospitals, and other overnight inpatient settings.

^‡Based on the patient’s zip code of residence during the fracture year as identified from the Medicare Master Beneficiary Summary.

After applying our model selection criteria, a 3-class model was estimated as the best fit for the prefracture data. Most older adults living with ADRD (n = 14 980/16 576, 90.3%) spent a majority of their prefracture DAH, which we distinguished as membership in a “Consistently High” DAH trajectory. However, 2 additional distinct trajectories, comprising nearly 10% of those with ADRD, were identified—one characterized by an initially similar but rapidly declining number of prefracture DAH (“Low and Decreasing”; n = 787/16 576, 4.7%) and the second with an initially lower but improving number of prefracture DAH (“Low but Increasing”; n = 809/16 576, 5.3%; Figure 1). Average posterior probabilities were 0.99, 0.89, and 0.91 for the Consistently High, Low but Increasing, and Low and Decreasing DAH trajectories, respectively, indicating appropriate model fit.

Figure 1.

Trajectories of days at home (DAH) before hip fracture among older adults living with Alzheimer’s disease or related dementia. The number of DAH each month ranged from 0 to 30. Dashed lines indicate identified trajectories; solid lines, predicted trajectories. The error bars represent 95% confidence intervals for the predicted number of days spent at home. The percentage of the overall sample in each trajectory is indicated in the figure legend. Average posterior probabilities were 0.99, 0.89, and 0.91 for the Consistently High, Low but Increasing, and Low and Decreasing DAH trajectories, respectively.

In the postfracture period, a model with 5 DAH trajectories had the best statistical fit (Figure 2). Two of the 5 DAH trajectories, representing the majority of survivors, were characterized as full recovery and categorized separately as rapid (“Rapid and Full Recovery”; n = 7 046/16 576, 42.5%) or delayed recovery (“Delayed but Full Recovery”; n = 4 598/16 576, 27.7%).

Figure 2.

Trajectories of days at home after hip fracture among older adults living with Alzheimer’s disease or related dementia. The number of days at home each month ranged from 0 to 30. Dashed lines indicate identified trajectories; solid lines, predicted trajectories. The error bars represent 95% confidence intervals for the predicted number of days spent at home. The percentage of the overall sample in each trajectory is indicated in the figure legend. Average posterior probabilities were 0.89, 0.89, 0.82, 0.80, and 0.89, respectively, for Persistently Low, Maladaptive Recovery, Recovery then Regression. Delayed but Full Recovery, and Rapid and Full Recovery.

However, less favorable postfracture DAH trajectories were observed (Figure 2), including (i) initial recovery but a subsequent and marked decline in DAH (“Recovery then Regression”; n = 1 239/16 576, 7.5%); (ii) slow and incomplete recovery (“Maladaptive Recovery”; n = 557/16 576, 3.4%); and (iii) a near complete lack of recovery characterized by spending nearly every day in the postfracture period hospitalized or in institutional care settings (“Persistently Low”; n = 3 136/16 576, 18.9%). The mean (SD) number of days spent at home in the 12 months after fracture ranged from 3.9 (11.3) for the Persistently Low DAH trajectory to 262.2 (122.9) for the Rapid and Full Recovery trajectory. The resultant model posterior probabilities were 0.89, 0.89, 0.82, 0.80, and 0.89, respectively, for Persistently Low, Maladaptive Recovery, Recovery then Regression, Delayed but Full Recovery, and Rapid and Full Recovery.

Prefracture trajectories were strongly associated with postfracture DAH trajectories (Supplementary Table 1). Notably, compared to the Consistently High trajectory, hip fracture survivors with ADRD in the Low and Decreasing trajectory prior to the fracture had more than double the probability of being in the Persistently Low postfracture DAH trajectory (39.1% vs 17.5%). Conversely, membership in a full-recovery postfracture trajectory was observed for 74% of those in the Consistently High preoperative trajectory, 44% of those in the Low but Increasing trajectory, and only 33% of those in the Low and Decreasing trajectory. These patterns were consistent in adjusted multinomial models (Figure 3).

Figure 3.

Adjusted transition probabilities conditional on prefracture trajectory. The multinomial model was adjusted for age, sex, hospital length of stay and intensive care unit use, Medicaid dual eligibility, Elixhauser score, surgical procedure type, days since dementia diagnosis, hospital teaching status, year of fracture, and county-level measures of poverty, nursing home bed availability, and rurality.

Prefracture DAH trajectories were also significantly associated with 1-year mortality rates. Unadjusted mortality at 1 year was 21% (n = 3 219/14 980) in the Consistently High trajectory, 26% (n = 213/809) in the Low but Increasing trajectory, and 32% (n = 254/787) for the Low and Decreasing trajectory. In models adjusted for the same covariates as the joint trajectories analysis, hip fracture survivors living with ADRD in the Low but Increasing preoperative DAH trajectory had a 27% higher risk for 1-year mortality (hazard ratio [HR] = 1.27, 95% CI 1.10–1.46) and those in the Low and Declining trajectory had a 65% higher risk for death at 1 year (HR = 1.65, 95% CI 1.45–1.87) as compared to those in the Consistently High trajectory (Figure 4).

Figure 4.

Cumulative incidence curve for 1-year mortality by prefracture trajectory. Cumulative incidence curves for 1-year mortality within each prefracture trajectory are depicted. Data were censored at 365 days.

We also observed notable demographic and clinical differences between the trajectory groups both before and after fracture (Supplementary Tables 2 and 3). The more unfavorable preoperative trajectory groups were more likely to include men, older adults who identified their racial or ethnic identity as Black or Hispanic, those dually eligible for Medicaid, and those with greater multimorbidity. Notable among the 5 postfracture subgroups, those in the Persistently Low trajectory were more likely to be eligible for Medicaid, more likely to identify as Black or African American, and more likely to live in suburban and isolated rural areas than those in full recovery trajectories.

Discussion

In this large national cohort of community-dwelling hip fracture survivors living with ADRD, we identified distinct trajectories of days spent at home before a hip fracture that were highly associated with postfracture DAH and 1-year mortality. Notably, those in the most unfavorable prefracture DAH trajectory had 65% higher risk for mortality at 1 year than those who spent the majority of their prefracture DAH. While only 1 in 10 older adults living with ADRD had unfavorable DAH trajectories prior to hip fracture, membership in these clinically distinct groups offered valuable prognostic information that could potentially inform surgical decision making and postdischarge care delivery. A second major finding in this work was the considerable variation in postfracture DAH among older adults with ADRD that likely represent resilience phenotypes. While older adults living with ADRD are often treated as a single clinical bloc following hip fracture (26), our findings suggest a common postoperative management pathway for all older adults with ADRD may not optimally meet the health needs and goals of this population.

Among the novel findings of our work were the clear delineations in postfracture outcomes based on distinct preoperative trajectories, heterogeneity that could be harnessed in a clinically meaningful way. Understanding the likelihood of recovery—or rehabilitation potential—among hip fracture survivors is critical for making discharge planning decisions (eg, inpatient rehabilitation vs skilled nursing facilities vs outpatient care) and initiating conversations about needs for supports such as durable medical equipment. For older adults with ADRD who may not be reliably able to report preoperative health, claims-based measures could offer valuable information to guide surgical decision making and discharge planning if integrated in real time into medical records (27). For example, identification of older adults with ADRD in a declining preoperative trajectory (those who were estimated to spend an average of just 3 DAH in our results) could facilitate conversations about, and more rapid referrals to, palliative care or hospice services when warranted—perhaps preventing patients with poor recovery potential from being “rehabbed to death” in post-acute care facilities and instead prompting discussion of alternative care options with seriously ill older adults that are more centered around goals of care (28). Conversely, membership in a Consistently High preoperative trajectory may indicate a patient may be a candidate for more aggressive restorative care such as inpatient rehabilitation that can provide up to 3 hours of therapy per day.

As personalized medicine becomes more ubiquitous in the management of older adults with ADRD, our findings might point toward one way to identify clinical subgroups of patients that may have distinct needs following hip fracture. Our findings also highlight a need to raise our recovery expectations for the majority of older adults with ADRD while vigilantly working to identify those who may need additional supports. Indeed, 2 in 3 hip fracture survivors with ADRD were predicted to have highly favorable recovery profiles—which helps undercut common misconceptions contributing to underutilization of aggressive rehabilitation care of this population (7). Conversely, there was a subgroup of 20% of hip fracture survivors with ADRD who spent nearly every day hospitalized or in an institutional setting, which our data suggested was driven by complex interactions between medical and social determinants of health. Notably, these patients were nearly twice as likely to be dually eligible for Medicaid as those in the Rapid and Full Recovery trajectory, 50% more likely to have been admitted to the ICU during the fracture admission, and 40% more likely live in a suburban or isolated rural area. This subgroup may have higher needs for both medical and social supports after discharge, which could be identified and addressed rapidly using preoperative claims data, and potentially acted upon during the hospital admission.

The strengths of this study include our use of nationally representative Medicare claims data with minimal missingness, and a large sample of older adults with ADRD. While prior work has generally concluded ADRD confers greater risk of poor outcomes after hip fracture (2), ours is the first to specifically disentangle pre- and postoperative recovery trajectories within this vulnerable subgroup. Our methods, implementing novel analytic strategies in claims data, could also be applied to other geriatric trauma populations. Nonetheless, there are several limitations of our work that should be noted. First, we did not include older adults covered by Medicare Advantage payers—meaning our findings cannot be generalized to this group. Second, we did not evaluate how interventions delivered during and after hospitalization may mediate or moderate recovery trajectories—an important consideration for future work. Third, while we adjusted our models for a robust array of covariates, there are several potentially important factors that are not captured in Medicare claims data, such as social supports or financial strain, that likely influence recovery. Additional limitations of administrative data include coarse information on chronic diseases (such as severity of cognitive impairment or presence of delirium) that could add important context to these results.

Hip fracture is a life-changing event for older adults, and represents a major contributor to future falls, disability, and death (29). Disability outcomes after hip fracture have not improved significantly over the past 2 decades, which may reflect poor tailoring of care (30). Our findings indicate claims-based measure of DAH prior to hip fracture is robustly associated with postfracture outcomes, and could inform decisions about tailored interventions for older adult with ADRD recovering from hip fracture.

Funding

J.R.F. was supported during the study by a Paul B. Beeson Emerging Leaders in Aging Award (K76AG074926) from the National Institute on Aging (NIA) and by the Maryland Claude D. Pepper Older Americans Independence Center of the NIA (P30 AG028747). J.M. was supported during the study by the Maryland Claude D. Pepper Older Americans Independence Center of the NIA (P30 AG028747). M.S. was supported by NIA grants RF1 NS128360, R03 AG070178, P30 AG028747-15S, and R01 AG048069. We also acknowledge the support of the University of Maryland, Baltimore, Institute for Clinical & Translational Research, and the National Center for Advancing Translational Sciences Clinical Translational Science Award (1UL1TR003098) in helping secure Medicare data and providing support for analysis.

Role of the Funders/Sponsor

The funders had no role in the design and conduct of the study, data collection, management, and analysis, and interpretation of the data, preparation, review, or approval of the manuscript, and decision to submit the manuscript for publication.

Conflict of Interest

J.R.F., M.S., and J.M. reported grants from the National Institute on Aging during the conduct of the study. J.M. serves on the Board of Directors for the Fragility Fracture Network and the American Orthopedic Association Multi-disciplinary Committee of the Own the Bone Program. The other authors declare no conflict.

Author Contributions

J.R.F. and C.C. had full access to all the data in the study and take responsibility for the integrity of the data and accuracy of the data analysis. Concept and design: J.R.F., C.C. Acquisition, analysis, or interpretation of the data: J.R.F., C.C., J.M., M.S., A.J., K.A.R., L.R. Drafting the manuscript: J.R.F. Critical revision of the manuscript (for important intellectual content): C.C., M.S., J.M., L.R. Statistical analysis: C.C., M.S., K.A.R., H.R. Obtained funding: J.R.F., M.S., J.M. Administrative, technical, material support: K.A.R., A.J., H.R. Supervision: J.M., M.S., C.C.