Factors that influence time to prosthesis receipt after lower limb amputation: A Cox proportional hazard model regression

Early mobility, functional independence, and ambulation are associated benefits after lower limb amputation (LLA), whereas an increased risk of clinical complications is associated with no prosthesis.


INTRODUCTION
Lower limb prosthetic rehabilitation aims to optimize functional mobility, independence, health, and quality of life (QoL). Mobility is a common challenge after lower limb amputation (LLA) despite the large amount of heterogeneity in patient phenotypes. 1,2 A prosthesis may help facilitate restoration of mobility in addition to increased independence in activities of daily living (ADL). 1,3 Independence with ADLs, safety within one's environment, and participation in social and workbased events are key factors that enhance QoL. 1,4 Early mobility, functional independence, and ambulation are associated with reduced health care use and economic burden whereas an increased risk of clinical complications is associated with no prosthesis. 5 In spite of these benefits, prostheses are not always provided to potential prosthetic candidates during the initial 12 months after amputation. 1,4 Individuals with LLA who do not receive a prosthesis have 2.6 times (95% confidence interval [CI]: 1.16-6.25) increased odds of mortality compared with those who receive a prosthesis. 6 Both 3-year morbidity and mortality rates remain high following amputation when controlling for comorbidities. 7 Yet, evidence suggests that early mobility, independence with ADLs, and ambulation promote good physical health while reducing mortality. 8 Currently available prosthetic technology provides expectation of an improvement in patient outcomes, including reduced mortality, morbidity, and avoidable unnecessary health care use. 5 A shorter time between LLA surgery and prosthesis provision may result in a more active and sustained recovery. 9,10 Previous studies have confirmed different benefits of shorter time to prosthesis but have been inconsistent with describing characteristics that influence the time between LLA and receipt of a prosthesis. In particular, the literature discussing the factors that may contribute to an earlier prosthesis provision within this population is relatively sparse. Amputations that are the result of dysvascular conditions or attributed to diabetes are common with a few studies noting dysvascular conditions contribute to a lower likelihood of prosthesis receipt. 8,11 Understanding factors that influence time from surgery to prosthesis can give health care providers insight to targeted interventions and allow improved care of those in need of a prosthesis.
The studies of rehabilitation outcomes that included prosthetic rehabilitation service were rarely able to account for a time component in their analysis. To a certain extent, it is perhaps a data source limitation. Publicly accessible databases, such as survey data (eg, National Health and Nutrition Examination Survey or Healthcare Cost and Utilization Project) are crosssectional and provide no obvious ability to conduct longitudinal analysis at the individual level. This is unfortunate as population-based data sets are a great resource for learning the characteristics associated with the outcomes that are more generalizable than the findings from restricted study settings (such as clinics, specific hospitals or systems, and single regions). Consideration of time to receipt of prosthesis has been instrumental to a number of researchers for identifying the rehabilitation outcomes and stratifying the individuals based on the associated factors. 12 Provision of a prosthesis begins with a prosthesis prescription during outpatient prosthetic rehabilitation and requires coordination across settings. 12 The provision and use of a prosthesis are critical components of a person's rehabilitation after LLA as they are associated with a person's ability to return to ADLs and reintegrate into social or work routines. 5,13 The timing from amputation surgery to initial device provision has several potential influences including the patient's age, cognitive ability, income, and rehabilitation setting. 12,13 However, there is no consensus on the factors that influence or predict provision of a prosthesis. By identifying factors that are associated with provision of a prosthetic device, it may be possible to have a better understanding of patient-level conditions (eg, diabetes status or gender), target modifiable patient factors (eg, mobility), standardize the prescription process, and better allocate resources in the future.
The goal of the current analysis was to determine factors that influence timing of prosthesis receipt among patients with LLA in a 12-month follow-up period and assess the median time to prosthesis receipt. It was hypothesized that timing of prosthesis receipt varies based on individual diabetes/vascular disease status, gender, and age. It was hypothesized that younger age, a more distal amputation level, and not having diabetes or vascular disease would increase likelihood of receiving a prosthesis in a shorter time.

Study population and data source
This retrospective cohort analysis used the International Business Machines (IBM) Corporation Watson/ Truven Health Analytics MarketScan (Watson) administrative database. The database contains de-identified records for commercial claims (billing data) representing approximately 25% of all commercials claims populated by payers in the United States. The full Watson database includes adjudicated claims that are aggregated into one data set including patient-level claims data from inpatient, outpatient, pharmacy, and enrollment history. A subset was extracted from the Watson database limited to only claims on patients who had a primary amputation from January 2014 through December 2016. Data is de-identified by IBM before release for analysis. Therefore, as these data are de-identified and comply with the Health Insurance Portability and Accountability Act, the subsequent analysis is not considered human subject research nor does it require approval from an institutional review board.
The final sample used for this analysis was limited to unique individuals 18 to 64 years of age who maintained continuous health coverage for the 3-year period (January 1, 2014 through December 31, 2016). Next, inclusion was based on all-cause amputation procedure (Appendix S1A); patients with initial major lower limb surgical amputation procedures (first procedure claim = index date) and no subsequent amputation surgery within the study period were included. Amputations at the level of partial foot or distal and hip disarticulation or proximal were excluded. The index period was set between March 2014-December 2015, allowing for and limiting analysis to 3 months of data before surgery and 12 months of data after surgery for all individuals. The final sample includes all eligible patients based on the stated criteria. If individuals did not meet these criteria they were not in the final analytic sample.

Study variables
The outcome of interest was time to prosthesis receipt over a 12-month window post LLA. Time to receipt of prosthesis (survival time) was calculated by determining the number of days from LLA surgery to the date the individual had a first claim for a lower limb prosthesis. The time variable was defined as the minimum of time to receipt of prosthesis or time to censorship. Prosthesis receipt was determined by presence of a lower limb prosthesis base code (ie, L-code) billed after LLA surgery (Appendix S1A). All types of prostheses were included and dichotomized; however, the specific kind of prosthesis was not extracted for analysis. Individuals who did not have a claim for a lower limb prosthesis during the 12-month follow-up period were considered a censored observation. Specifically, time to censoring was defined as the number of days from amputation surgery to 12 months post surgery.
Age was treated as a continuous variable, years 18 to 64. Individuals were classified as either having diabetes mellitus type II or vascular disease (Appendix S1A) based on presence of diagnosis codes at baseline. Additional variables included in the model were gender (binary) and amputation level categorized to below knee (BK) or above knee (AK), which was determined by LLA surgical procedure. All covariates were treated with the assumption to persist throughout the study period.

Data analysis
Descriptive statistics were used to summarize the characteristics of the cohorts with chi-square or Student's ttest applied to discern group differences. The bivariate survivorship function [S(t) = Prob (time to prosthesis is T A B L E 1 Sample demographic characteristics and inferential statistics (chi-square test and t-test) stratified by receipt of prosthesis longer than t)] and percentiles were estimated using the Kaplan-Meier (KM) product-limit survival distribution. KM is a nonparametric approach. A log-rank test was applied to determine if the survival function differed based on groups. KM estimator for survival function facilitated the calculation of median days to prosthesis receipt and its 95% pointwise confidence interval (CI). Survival function in this study was interpreted as the cumulative probability or rate of receiving a prosthesis within the 12 months assessed. The final adjusted association between successful prosthesis receipt and the independent variables was quantified through multivariate analysis using a Cox proportional hazards (CPH) model. After KM analysis variables identified with a log-rank p value ≤.05 were entered into the multivariate CPH regression model. The results were reported as hazard ratios (HR) and their 95% CI based on the conditional estimates of the model. Model diagnostics were conducted including testing the proportional hazards assumption using the function cox.zph() to assess each covariate and a global p value using Schoenfeld residuals against time from the survival package. 14 A p value greater than .05 means the proportional hazards assumption is not violated for a CPH regression model fit; the global Schoenfeld for the study data p value was .12. The average estimated rate of prosthesis receipt while adjusting for covariates was extracted from the CPH model. All data management and analyses were conducted using R studio version 1.

RESULTS
Among the 510 adults who maintained continuous coverage, 87% received a prosthesis within 12 months (365 days) post amputation surgery with an average time to prosthesis receipt of 130 days (Table 1). Of those who received a prosthesis, 352 had a BK amputation, whereas 91 had an AK amputation. In contrast, those who did not receive a prosthesis within 12 months were more evenly distributed, 33 with BK amputation and 34 with AK amputation. Overall, the sample consisted of individuals with an average age of 52 years, and 70% were male (Table 1). Univariate analysis of prosthesis receipt was carried out with KM models, stratified by patient characteristics; comparison of survival curves was assessed using the log-rank test. The median survival, or number of days to prosthesis receipt, with the 95% CIs and p value for the log-rank test results are presented (Table 2) along with each KM curve stratified by specific patient characteristic ( Figure 1). Diabetes/vascular status was significantly associated with time to prosthesis receipt, such that amputees with diabetes/vascular disease tended to receive their prostheses earlier than those without diabetes/vascular disease (χ 2 = 5.5, p = .02). The median survival time to prosthesis receipt for those with diabetes or vascular disease was 113 (95% CI: 101-129) days compared to 124 (95% CI: 113-159) days for those without diabetes or vascular disease ( Figure 1A). Also, time to receipt of prosthesis was associated with amputation level (χ 2 = 27.5, p = <.001). The median survival time for an individual with a BK amputation to receive a prosthesis was 107 (95% CI: 102-119) days, compared to 168 (95% CI: 137-204) days for someone with an AK amputation ( Figure 1B). Timing of prosthesis receipt was significantly different based on gender (χ 2 = 10.5, p = .001), 141 (95% CI: 126-162) days for women compared to 106 (95% CI: 96-119) for men ( Figure 1C). Region of care was not associated with time to prosthesis receipt (χ 2 = 0.8, p = .9) and therefore not included in the final Cox adjusted regression model. The CPH regression model is useful for examining the characteristics against the adjusted probability of prosthesis receipt over time and the final adjusted model is visualized in a adjusted survival curve for the full sample ( Figure 2). The unadjusted and adjusted HRs, or probability of receiving a prosthesis at a given time post LLA while controlling for covariates, are presented in Table 3. Those with an amputation level of BK had an increased probability of receiving a prosthesis sooner with the adjusted HR of 1.79 (95% CI: 1.42-2.26) as compared to those with AK amputation. Those who were male were more likely to receive a prosthesis F I G U R E 2 Final adjusted Cox proportional hazard model, controlling for amputation level, diabetes/vascular status, gender, and age. The ordinate represents the probability of an individual without a prosthesis; for example, at time of amputation surgery (t = 0) 100% of individuals did not have a prosthesis. Women are represented by the grey line who received a prosthesis on average later compared to men (black line) while controlling for covariates. The average rate or when 50% of sample received a prosthesis was 138 days (95% confidence interval: 113-185) T A B L E 3 Cox proportional hazard ratio (HR) unadjusted and adjusted model results of lower limb amputation surgery to receipt of prosthesis. A hazard ratio greater than 1 represents an increased likelihood or rate of prosthesis receipt within 12 months post amputation

DISCUSSION
In this retrospective cohort analysis, 443 individuals received a prosthesis within 12 months of LLA (79% BK and 21% AK). This statistic includes only patients who were continuously enrolled in the same commercial plan for the entire study period and therefore does not represent those who may have switched plans or expired after LLA. In this real-world analysis of commercial claims data, the unadjusted median survival time between LLA surgery and prosthesis receipt is 107 days for BK amputation and 168 days for AK amputation. The overall adjusted average rate estimate of time to prosthesis receipt is 138 days. The hypothesis was partially confirmed with the finding that amputation level influenced timing of prosthesis receipt, age did not significantly influence timing of receipt, and although diabetes/vascular status influenced timing, it did so contrary to expectation. It was found that those with diabetes or vascular disease were 22% more likely to receive a prosthesis earlier, potentially suggesting that there are underlying differences beyond diabetes or vascular disease. Perhaps those who have an amputation not attributed to diabetes or vascular disease are experiencing more complicated amputations secondary to trauma. Alternatively, this may be a limitation of the data set as individuals with nondiabetic or vascular amputation may have changed enrollment with their provider (eg, loss of employment) and subsequently would have been excluded from analysis. As expected, individuals with a BK amputation had an 80% increased probability (HR: 1.8, 95% CI: 1.42-2.26) to receive a prosthesis earlier than someone with an AK amputation within 12 months. This appears consistent with similar literature that reported those with BK amputation often have a higher functional status post amputation, 6,11,15 although previous work has also demonstrated that premorbid functional and ambulatory status is a greater predictor of successful prosthetic outcomes. 16 This study reflects a prosthesis delivery rate higher than others, such as by Fletcher et al. 17 , yet it is difficult to compare to this study because their population was only elderly patients (greater than 65) who all had an LLA due to vascular disease. In contrast, the current finding of almost 87% receiving a prosthesis within 12 months of LLA surgery, comprises younger (under 65) individuals with both vascular and nonvascular causes of amputation who maintain the same commercial plan for the study period. The limitation of only individuals who maintained their insurance coverage heavily skews this sample in terms of prosthesis receipt because someone with insurance is more likely to receive care compared to someone who is uninsured.
There was also an association with gender as women tend to have a slightly longer wait (approximately 1 month difference) for prosthesis receipt. This may explain the literature that notes women tend to have worse functional outcomes after LLA as they are potentially not being put upright to regain mobility as quickly. [18][19][20] This is not entirely different from other studies that have noted gender disparities in functional outcomes. For example, such differences have been noted among stroke survivors 21 and post-total knee replacement. 22 One previous study evaluating factors that influenced prosthesis receipt noted gender was not significantly associated with receipt. 12 In comparison, however, that analysis evaluated only individuals with AK amputation, was a smaller sample size, limited to a single county's data, and did not investigate impact on timing of prosthesis receipt. Gender differences appear to be multifactorial; however, there should be further investigation on gender-based disparities among those with LLA.
Previous research has suggested that chronic conditions, such as diabetes and vascular disease, are associated with increased likelihood of LLA and those with diabetes have an increased risk of infection, falling, and other complications that may delay prosthesis provision. 23,24 The current results suggest that the comorbid conditions of diabetes and/or vascular disease do not prevent prosthesis receipt as visually presented with the Kaplan-Meier curve (Figure 2, p = .02). Additionally, as supported by previous work, diabetes or vascular disease alone do not prevent successful prosthesis use. 25 This study has several strengths, including its large sample size, and it is based on a nationally representative sample of commercial claims. Also, this study focuses on a specific cohort of relatively younger, working-age adults continuously enrolled in the same commercial insurance plan, which adds to previous work limited from being heavily skewed toward inclusion of older adults.
There are limitations to this study as well. First, administrative data do not provide the exact cause of amputation, and therefore, this sample includes allcause LLA. It is also not possible to differentiate potential clinical reasons, provider workflows, or specific payer plan policies that may affect timing for an individual to receive a prosthesis. It is possible that a delayed fitting may be associated with other health complications, administrative (nonpatient) issues, lack of social support, or another unseen complication that contributes to adverse events thus increasing time from surgery to prosthesis receipt. For example, it has been suggested that lack of social support or marital status (ie, being single) may negatively affect prosthetic rehabilitation. 6,17 Future studies could take these factors into account.
It is worth acknowledging the specific population analyzed in this study. This sample includes all-cause amputation with the intention to represent usual clinical practice among those who are commercially insured, who consequently tend to be younger and perhaps experience more trauma-related injuries than an older population. A disease-specific cohort among those commercially insured may not be the same as a cohort of those over 65 years with different insurance options, such as Medicare or those of any age but who use the Veterans Health Administration for prosthetic services. Additionally, a commercially insured cohort may not represent those who are covered by Medicaid or uninsured. This sample population consists of all primary, major LLA cases from across the United States who maintained continuous enrollment over a 3-year period. This group, therefore, may better reflect clinical practice across all regions of the United States for those who maintain commercial (private) insurance plans. However, it is expected that a percentage of individuals may not receive a prosthesis owing to not maintaining enrollment. The current analysis did not have visibility into these individuals. It should be noted it is not clear if dropping or switching enrollment to another insurance provider precipitates not getting a prosthesis or if issues with getting a prosthesis lead to changing insurance provider.

CONCLUSION
Despite the limitations in this study, the results are important because of their practical implications. This research expands the understanding of factors that influence the likelihood of receiving a prosthesis along with the timing of prosthesis receipt after LLA. The fact that all individuals remained enrolled in their insurance plan for the study period allows for a unique perspective evaluating a younger cohort who may reflect clinical practice for prosthetists. The delivery of an initial prosthesis may have significant impact on an individual's future recovery and rehabilitation performance especially among working-age adults as return to work is often an important goal. Among this sample of working-age adults with commercial insurance, it appears that within 5 months or less at least half of the population receives a prosthesis while controlling for covariates; however disparities in timing and access to a prosthesis based on amputation level and gender should be addressed.