Abstract
Background
We sought to determine whether several pre-operative socioeconomic status (SES) variables meaningfully improve predictive models for primary total knee arthroplasty (TKA) length of stay (LOS), facility discharge, and clinically significant Veterans RAND-12 Physical Component Score (PCS) improvement.
Methods
We prospectively collected clinical data on 2,198 TKAs at a high-volume rural tertiary academic hospital from April 2011 through March 2016. SES variables included race/ethnicity, living alone, education, employment, and household income, along with numerous adjusting variables. We determined individual SES predictors and whether the inclusion of all SES variables contributed to each 10-fold cross-validated area under the model’s area under the receiver operating characteristic (AUC). We also used 1000-fold bootstrapping methods to determine whether the SES and non-SES models were statistically different from each other.
Results
At least 1 SES predicted each outcome. Ethnic minority patients and those with incomes<$35,000 predicted longer LOS. Ethnic minority patients, the unemployed, and those living alone predicted facility discharge. Unemployed patients were less likely to achieve PCS improvement. Without the 5 SES variables, the AUC values of the LOS, discharge, and PCS models were 0.74 (95% CI 0.72–0.77, “acceptable”; 0.86 (CI 0.84–0.87, “excellent”); and 0.80 (CI 0.78–0.82, “excellent”), respectively. Including the 5 SES variables, the ten-fold cross-validated and bootstrapped AUC values were 0.76 (CI 0.74–0.79); 0.87 (CI 0.85–0.88); and 0.81 (0.79–0.83), respectively.
Conclusions
We developed validated predictive models for outcomes after TKA. Although inclusion of multiple SES variables provided statistical predictive value in our models, the amount of improvement may not be clinically meaningful.
Keywords: Total knee arthroplasty, socioeconomic status, length of stay, discharge disposition, patient-reported outcomes, predictive models
Introduction
Total knee arthroplasty (TKA) is an established cost-effective surgical procedure that is designed to relieve knee pain and disability.1,2 From 2001 through 2011, TKA incidence increased 93% to 718,000 cases annually in the United States, and now has the third highest incidence of any surgical procedure.3 In 2011, TKA had the second-highest aggregate costs for any surgery at $11.3 billion.3 Due to its popularity, high cost, and lagging reimbursement rate,4,5 TKA has been subjected to significant cost containment efforts by hospitals, insurers, and the government.6 Despite this, high variations remain in TKA outcomes among U.S. hospitals, with potentially immense influence on patient quality of life, quality of care, and healthcare costs.7
Patient-reported outcomes (PROs) and quality reporting are evolving rapidly in medicine. However, these collection efforts may come at significant cost, time commitment, and data analysis: current estimates indicate that collection of external quality measures takes approximately 11 practice hours per orthopaedic surgeon per week (1 hour specifically by the physician) and approximately $31,500 per orthopaedic surgeon per year.8 The Comprehensive Care for Joint Replacement (CJR) will likely increase these commitments further. These efforts may create large collection burdens for both practices and patients; for example, the 2,060 patients (2,198 TKAs) discussed in this article have given 1,314,420 answers to 2,238 different survey questions posed by our institution from April 2011 through March 2016.
Several preoperative factors have been noted to be associated with TKA outcomes, including gender,9–12 age,9–13 physical function,9,12–14 mental function,15 body mass index,12,16 and medical comorbidities.12,17 Little is known about how various demographic and socioeconomic status (SES) variables are associated with TKA outcomes after adjustment for other factors, although they are studied regularly across other medical disciplines.18 Until recently, SES variables were rarely studied in Orthopaedics, particularly in rural areas for elective surgeries after adjusting for clinical, PROs, and multiple SES variables.19–22 Even among these new reports, none used more than 1 self-reported SES measure in conjunction with patient-reported outcomes, nor provided comparative predictive models for whether SES variables independently contributed to models. It is possible that adding SES variables to TKA predictive models may offer diminishing returns and little contributory value.23 Apart from race and ethnicity, SES variables are not collected currently in the American Joint Replacement Registry24 or the Function and Outcomes Research for Comparative Effectiveness in Total Joint Replacement consortium (FORCE-TJR registry).25
We sought to determine whether five self-reported SES measures (race/ethnicity, living alone, education, employment, and household income) contribute independently and meaningfully to TKA outcome predictive models in a rural area, including length of stay (LOS) beyond 3 days, facility discharge, and clinically significant patient-reported improvement of physical function. We elected to only use preoperative variables that may be self-reported by the patient outside of a medical setting; other papers have noted the ability to create strong predictive models using only these types of variables.15,23 We hypothesized that SES variables would predict each outcome in an adjusted multivariate model, and that the addition of SES variables would significantly improve each predictive model.
Materials and Methods
Setting
Our study uses data from our prospective Orthopaedic Operational Data Repository, at a rural tertiary academic medical center in northern New England, from April 2011 through March 2016. All variables are collected prospectively through the electronic medical record (EMR). Data collection was built into normal daily practice for all patients. Our institutional review board waived the requirement for individual informed consent. There were no exclusions among completed primary TKA procedures. The institutional arthroplasty clinical pathway went through a standardization process in April 2011 (concurrent with our EMR implementation) and did not vary significantly during the research time period.
Variables
The selected self-reported SES variables included race/ethnicity [non-Hispanic White (“white”) or not (“ethnic minority”: consisting of Asian, Black, Hispanic, and/or Native American), reflecting the demographics of rural New England; due to low counts, it would not be valid to investigate each race/ethnicity group separately], living alone (no, yes), completed education (postgraduate, college, some college, high school or less), employment (working/student, retired for non-health reasons, not working), and household income (more than $75,000, $50,000–75,000, $35,000–50,000, less than $35,000, and refused to answer).26 We used the value closest to date of surgery for SES variables when captured multiple times for the same patient.
Among the available variables, we focused on the following preoperative adjusters: age,12 gender,12 surgeon,12 alcohol use,18 tobacco use,18,27 Charlson Comorbidity Score,12 Veterans RAND-12 (VR-12) physical component score (PCS),12,28–30 VR-12 mental component score (MCS),31 year of surgery, bilaterality,32 and BMI.12,16 Additionally, the PCS improvement model also includes an adjusting variable for the time period in which postoperative PCS was captured. Our PCS preoperative, postoperative, and both time period capture percentages are 92%, 79%, and 78%, respectively, which generally meet the PRO capture rates in the well-known AJRR, FORCE-TJR, and California Joint Replacement Registry (CJRR) registries.33 Table 1 shows counts of each variable in our data set. We included SES missing values as their own category and retained them in all analyses and tables to determine whether those who did not answer or were not asked SES questions varied from patients who did. Only the SES variables had significant missingness. Missing values for other variables are included in the models and described in Table 1, but are not displayed.
Table 1.
Counts of relevant variables within the institutional total knee arthroplasty repository (n=2,198 surgeries among 2,060 individuals)
| Variable | Counts, N = 2,198 | % |
|---|---|---|
| Race (ref = Non-Hispanic White) | 2,153 | 98 |
| Ethnic Minority | 45 | 2 |
| Living Alone (ref = No) | 483 | 22 |
| Yes | 162 | 7 |
| Missing | 1,553 | 71 |
| Education (ref = Postgrad) | 183 | 8 |
| College | 165 | 8 |
| Some College | 190 | 9 |
| High School or Less | 224 | 10 |
| Missing | 1,436 | 65 |
| Employment (ref = Working/Student) | 254 | 12 |
| Retired, Not Health Related | 276 | 13 |
| Not Working | 232 | 11 |
| Missing | 1,436 | 65 |
| Household Annual Income (ref = $75,000+) | 124 | 6 |
| $50–75,000 | 92 | 4 |
| $35–50,000 | 72 | 3 |
| <$35,000 | 128 | 6 |
| Refused to Answer | 150 | 7 |
| Missing | 1,632 | 74 |
| Surgeon (ref = Surgeon 1) | 541 | 25 |
| Surgeon 2 | 363 | 17 |
| Surgeon 3 | 29 | 1 |
| Surgeon 4 | 4 | 0.2 |
| Surgeon 5 | 51 | 2 |
| Surgeon 6 | 358 | 16 |
| Surgeon 7 | 425 | 19 |
| Surgeon 8 | 109 | 5 |
| Surgeon 9 | 80 | 4 |
| Surgeon 10 | 238 | 11 |
| Age Group (ref = <55) | 301 | 14 |
| 55–59 | 323 | 15 |
| 60–64 | 399 | 18 |
| 65–69 | 462 | 21 |
| 70–74 | 328 | 15 |
| 75–79 | 223 | 10 |
| 80+ | 162 | 7 |
| Sex (ref = Male) | 938 | 43 |
| Female | 1,260 | 57 |
| Preoperative Alcohol Use (ref = No) | 804 | 38 |
| Yes | 1,305 | 62 |
| Preoperative Tobacco Use (ref = Never) | 1,009 | 47 |
| Quit | 999 | 46 |
| Yes | 151 | 7 |
| Charlson Score (ref = 0) | 1,284 | 58 |
| 1 | 456 | 21 |
| 2+ | 458 | 21 |
| PCS Preoperative Mean (SD, Range), n = 2,012 | 31.2 (11.0, 6.9 – 70.4) | |
| 50+ | 138 | 7 |
| 40–49.99 | 301 | 15 |
| 30–39.99 | 479 | 24 |
| 20–29.99 | 802 | 40 |
| <20 | 292 | 15 |
| VR-12 MCS Preoperative Mean (SD, Range) | 53.4 (13.2, 18.4 – 75.1) | |
| 60+ | 844 | 42 |
| 50–59.99 | 469 | 23 |
| 40–49.99 | 301 | 15 |
| 30–39.99 | 257 | 13 |
| <30 | 141 | 7 |
| VR-12 PCS Post-Op Mean (SD, Range), n = 1,812 | 41.3 (12.5, 8.5 – 64.8) | |
| VR-12 PCS Post-Op Time Period (ref = 0 – 45 Days Post-Operative) a | 268 | 16 |
| 46 – 299 Days Post-Operative | 512 | 30 |
| 300 – 420 Days Post-Operative (1 Year) | 720 | 42 |
| 421 + Days Post-Operative | 225 | 13 |
| Year (ref = April – December 2011) | 321 | 15 |
| 2012 | 505 | 23 |
| 2013 | 522 | 24 |
| 2014 | 403 | 18 |
| 2015 | 368 | 17 |
| January – March 2016 | 79 | 4 |
| Bilateral TKA (ref = No) | 1,687 | 77 |
| Yes | 511 | 23 |
| BMI Preoperative Mean (SD, Range) b | 32.4 (7.4, 16.0 – 70.4) | |
| Normal, <25 | 279 | 14 |
| Overweight, 25–29.99 | 602 | 30 |
| Obese, 30–34.99 | 488 | 24 |
| Severe Obese, 35–39.99 | 345 | 17 |
| Morbid Obese, 40+ | 290 | 14 |
| Length of Stay, Days Mean (SD, Range) | 3.0 (1.5, 1 – 30) | |
| 1 | 140 | 6 |
| 2 | 680 | 31 |
| 3 | 901 | 41 |
| 4 | 291 | 13 |
| 5 | 97 | 4 |
| 6 | 42 | 2 |
| 7 | 19 | 1 |
| 8 | 9 | 0 |
| 9 | 8 | 0 |
| 11 | 2 | 0 |
| 12 | 2 | 0 |
| 13 | 2 | 0 |
| 14 | 2 | 0 |
| 15 | 1 | 0 |
| 21 | 1 | 0 |
| 30 | 1 | 0 |
| Length of Stay (ref <4 Days) | 1,721 | 78 |
| > 3 Days | 477 | 22 |
| Discharge Disposition | ||
| Home | 1,364 | 62 |
| Facility | 832 | 38 |
| PCS Improvement Mean (SD, Range) | 10.1 (13.9, −39.5 – 44.5) | |
| PCS Clinically Significant Improvement (VR-12 Post-Operative – VR-12 Pre-Operative), >5 score increase (ref = No) | 642 | 37 |
| Yes | 1,083 | 63 |
Some additional non-SES variables have minor missingness for all tables and models. These surgeries contained categories of missing and were included in the models, but are not displayed. Their counts include postoperative VR12 PCS (n=386, 18%), preoperative BMI (n=194, 9%), preoperative VR12 PCS (186, 8%), preoperative VR12 MCS (186, 8%), preoperative alcohol use (89, 4%), preoperative tobacco use (31, 1%), and discharge disposition (2 hospital deaths, 0%). Tobacco use also included a measure of “Passive”, which only had 8 surgeries; this value was retained in the models but is not displayed.
Only includes post-operative scores among surgeries that also had a pre-operative score. Of the 2,198 surgeries, 2,012 (91%) had a pre-operative VR-12, 1,812 (82%) had a post-operative VR-12, and 1,725 (79%) had both pre-operative and postoperative.
BMI was calculated from height and weight measurements collected within clinic visits and were not self-reported.
Models
We included models for predicting three different outcomes for TKA: length of stay over 3 days (LOS, no/yes),26 facility discharge (no/yes) (Table 2),12 and clinically significant PCS improvement, determined as an improvement of at least 5 points on the normalized 0 – 100 scale (no/yes).15,30 We used dichotomous outcomes to enable satisfactory predictive models. The change score was measured as the latest postoperative recorded PCS subtracted by the latest preoperative score. As this model necessitates completeness of scores from both time periods, only 1,723 TKAs (78.4%) were included in this model. All surgeries were used for the LOS and discharge models.
Table 2.
Definition and counts of home and inpatient discharge, as reported in EMR.
| Discharge Type | Count | % | Discharge Category |
|---|---|---|---|
| Custodial Care | 1 | 0.1 | Home |
| Deceased before discharge | 2 | 0.1 | -- |
| Home | 60 | 2.7 | Home |
| Home with Visiting Nurse | 1,303 | 59.3 | Home |
| Intermediate Care Facility | 1 | 0.1 | Facility |
| Other Short Term General Hospital | 1 | 0.1 | Facility |
| Psychiatric Hospital – Stand Alone | 1 | 0.1 | Facility |
| Rehabilitation Center in a Facility | 15 | 0.7 | Facility |
| Rehabilitation Center –Acute Care | 209 | 9.5 | Facility |
| Rehabilitation Center –Stand Alone | 91 | 4.1 | Facility |
| Skilled Nursing Facility | 244 | 11.1 | Facility |
| Swing Bed | 270 | 12.3 | Facility |
| Total | 2,198 | 100 | Home 1,364 (62.1%), Facility 832 (37.9%) |
Analyses
All analyses were conducted using Stata MP12.34 We used multivariate logistic regression models for all outcomes. To evaluate each model’s ability to distinguish between the dichotomous outcomes, we determined the area under the receiver operating characteristic (AUC) curve with 10-fold cross-validation to estimate the AUC in sub-samples of our data.23,35 An AUC from 0.70 to 0.80 is acceptable and 0.80 to 0.90 is excellent.23,36 Models with and without the SES variables (race/ethnicity, living alone, education, employment, and household income) were both documented to determine whether the AUC cross-validation measurements improved (higher on scale of 0 – 1) with the SES variables. In addition to cross-validation, we also performed 1000-fold bootstrapping techniques to statistically compare the AUCs with and without SES variables.37 All models clustered on the individual patient to account for separate contralateral primary TKAs for 138 patients and used robust standard errors to account for the observational nature of the study.
Results
Sample Characteristics
In the mentioned time period, the Department of Orthopaedics performed 2,198 primary TKAs among 2,060 individual patients. Of the surgeries, the patient mean age was 65.6 (SD 9.8, Range 26–90), 57.3% were female, 98.0% were white, and mean preoperative body mass index (BMI) was 32.4 (SD 7.4, range 16.0–70.4). Table 1 includes basic counts and categories of our variables of interest and adjusters.
Model 1: LOS over 3 days
Among our 2,198 surgeries, 22% had a LOS over 3 days (Table 1). Among the SES variables, both ethnic minority patients and those with annual household incomes under $35,000 predicted longer LOS (ORs 2.39 and 2.74, P=0.003 and P=0.017, respectively). Living alone and “not working” trended towards longer LOS (P=0.074 and P=0.058, respectively). Surgeon, age ≥75, preoperative Charlson score higher than 0, patients with lower MCS, year of surgery, bilateral TKAs, and morbid obesity predicted higher odds of longer LOS (Table 3). Lower preoperative PCS scores strongly predicted longer LOS in a dose-response fashion. Alcohol use predicted against longer LOS. Without the 5 SES variables, the LOS predictive model had a ten-fold cross-validated and bootstrapped AUC of 0.74 (CI 0.72–0.77), indicating an acceptable predictive value;23,35 adding the 5 SES variables slightly improved the AUC to 0.76 (CI 0.74–0.79), which was also considered acceptable. The difference between the AUCs was statistically significant (P<0.0001).
Table 3.
Multivariate logistic model for whether separate SES variables individually predict LOS>3 days (22%) among primary TKAs (n=2,194)
| With SES Variables | No SES Variables | |||||
|---|---|---|---|---|---|---|
| Variable | OR | 95% CI | P-Value | OR | 95% CI | P-Value |
| Race (ref = Non-Hispanic White) | ||||||
| Other | 2.39 | 1.35–4.25 | 0.003 | -- | -- | -- |
| Living Alone (ref = No) | ||||||
| Yes | 1.60 | 0.96–2.67 | 0.074 | -- | -- | -- |
| Missing | 1.42 | 0.69–2.89 | 0.340 | -- | -- | -- |
| Education (ref = Postgrad) | ||||||
| College | 1.51 | 0.79–2.88 | 0.216 | -- | -- | -- |
| Some College | 1.19 | 0.62–2.31 | 0.601 | -- | -- | -- |
| High School or Less | 1.29 | 0.68–2.44 | 0.430 | -- | -- | -- |
| Missing | 1.00 | 0.30–3.28 | 0.996 | -- | -- | -- |
| Employment (ref = Working/Student) | ||||||
| Retired, Not for Health | 0.81 | 0.47–1.39 | 0.440 | -- | -- | -- |
| Not Working | 1.65 | 0.98–2.77 | 0.058 | -- | -- | -- |
| Missing | 1.24 | 0.40–3.82 | 0.709 | -- | -- | -- |
| Household Annual Income (ref = $75,000+) | ||||||
| $50–75,000 | 2.16 | 0.88–5.32 | 0.093 | -- | -- | -- |
| $35–50,000 | 0.78 | 0.28–2.20 | 0.643 | -- | -- | -- |
| <$35,000 | 2.74 | 1.20–6.25 | 0.017 | -- | -- | -- |
| Refused to Answer | 1.15 | 0.51–2.62 | 0.732 | -- | -- | -- |
| Missing | 1.99 | 0.80–4.91 | 0.137 | -- | -- | -- |
| Surgeon (ref = Surgeon 1) | ||||||
| Surgeon 2 | 2.60 | 1.76–3.84 | <0.001 | 2.61 | 1.78–3.83 | <0.001 |
| Surgeon 3 | 2.64 | 0.90–7.73 | 0.077 | 2.48 | 0.86–7.12 | 0.091 |
| Surgeon 4* | -- | -- | -- | -- | -- | -- |
| Surgeon 5 | 1.44 | 0.56–3.71 | 0.449 | 1.45 | 0.58–3.61 | 0.426 |
| Surgeon 6 | 1.82 | 1.23–2.70 | 0.003 | 1.82 | 1.23–2.70 | 0.003 |
| Surgeon 7 | 2.62 | 1.81–3.78 | <0.001 | 2.65 | 1.85–3.81 | <0.001 |
| Surgeon 8 | 1.06 | 0.56–2.03 | 0.857 | 1.14 | 0.61–2.15 | 0.675 |
| Surgeon 9 | 0.49 | 0.17–1.40 | 0.183 | 0.52 | 0.18–1.48 | 0.218 |
| Surgeon 10 | 1.73 | 1.10–2.73 | 0.018 | 1.70 | 1.08–2.67 | 0.022 |
| Age Group (ref = <55) | ||||||
| 55–59 | 0.95 | 0.62–1.45 | 0.800 | 0.91 | 0.60–1.39 | 0.670 |
| 60–64 | 0.90 | 0.60–1.36 | 0.629 | 0.88 | 0.59–1.32 | 0.538 |
| 65–69 | 1.20 | 0.82–1.78 | 0.350 | 1.17 | 0.80–1.72 | 0.427 |
| 70–74 | 1.29 | 0.83–2.02 | 0.260 | 1.22 | 0.79–1.87 | 0.375 |
| 75–79 | 2.11 | 1.32–3.36 | 0.002 | 1.89 | 1.21–2.97 | 0.005 |
| 80+ | 2.58 | 1.56–4.26 | <0.001 | 2.38 | 1.45–3.89 | 0.001 |
| Sex (ref = Male) | ||||||
| Female | 1.20 | 0.95–1.53 | 0.123 | 1.24 | 0.98–1.56 | 0.072 |
| Preoperative Alcohol Use (ref = No) | ||||||
| Yes | 0.71 | 0.55–0.90 | 0.004 | 0.66 | 0.52–0.84 | 0.001 |
| Preoperative Tobacco Use (ref = Never) | ||||||
| Quit | 1.18 | 0.93–1.49 | 0.176 | 1.20 | 0.95–1.52 | 0.127 |
| Yes | 1.15 | 0.70–1.88 | 0.577 | 1.29 | 0.80–2.09 | 0.296 |
| Charlson Score (ref = 0) | ||||||
| 1 | 1.41 | 1.06–1.86 | 0.016 | 1.38 | 1.05–1.84 | 0.021 |
| 2+ | 1.64 | 1.22–2.20 | 0.001 | 1.62 | 1.22–2.17 | 0.001 |
| VR-12 PCS Preoperative (ref = 50+) | ||||||
| 40–49.99 | 2.43 | 1.17–5.01 | 0.017 | 2.62 | 1.28–5.36 | 0.008 |
| 30–39.99 | 2.99 | 1.49–5.96 | 0.002 | 3.20 | 1.62–6.32 | 0.001 |
| 20–29.99 | 2.94 | 1.48–5.85 | 0.002 | 2.17 | 1.62–6.23 | 0.001 |
| <20 | 4.55 | 2.18–9.46 | <0.001 | 4.93 | 2.41–10.09 | <0.001 |
| VR-12 MCS Preoperative (ref = 60+) | ||||||
| 50–59.99 | 1.17 | 0.85–1.60 | 0.328 | 1.23 | 0.90–1.67 | 0.198 |
| 40–49.99 | 1.53 | 1.08–2.17 | 0.017 | 1.63 | 1.15–2.31 | 0.006 |
| 30–39.99 | 1.94 | 1.32–2.84 | 0.001 | 2.10 | 1.44–3.07 | <0.001 |
| <30 | 1.30 | 0.76–2.23 | 0.342 | 1.36 | 0.80–2.30 | 0.255 |
| Year (ref = 2011) | ||||||
| 2012 | 0.61 | 0.42–0.91 | 0.015 | 0.64 | 0.44–0.95 | 0.026 |
| 2013 | 0.73 | 0.49–1.09 | 0.123 | 0.66 | 0.45–0.96 | 0.032 |
| 2014 | 0.80 | 0.51–1.25 | 0.324 | 0.75 | 0.48–1.15 | 0.189 |
| 2015 | 0.47 | 0.28–0.78 | 0.004 | 0.41 | 0.25–0.68 | 0.012 |
| 2016 | 0.37 | 0.13–1.03 | 0.055 | 0.36 | 0.14–0.97 | 0.042 |
| Bilateral TKA (ref = No) | ||||||
| Yes | 2.53 | 1.96–3.27 | <0.001 | 2.39 | 1.86–3.08 | <0.001 |
| BMI Preoperative (ref = Normal, <25) | ||||||
| Overweight, 25–29.99 | 0.94 | 0.63–1.39 | 0.741 | 0.91 | 0.62–1.35 | 0.648 |
| Obese, 30–34.99 | 1.10 | 0.73–1.65 | 0.641 | 1.11 | 0.75–1.65 | 0.597 |
| Severe Obese, 35–39.99 | 1.28 | 0.83–1.98 | 0.258 | 1.32 | 0.86–2.02 | 0.200 |
| Morbid Obese, 40+ | 1.61 | 1.02–2.55 | 0.040 | 1.64 | 1.04–2.56 | 0.031 |
| 10-fold cross-validation AUC | ||||||
| 0.76 | 0.74–0.79 | 0.74 | 0.72–0.77 | |||
Clustered on 2,057 patients and using robust standard errors to account for observational data. Surgeon 4 performed 4 surgeries in this time period and they were excluded from this analysis for lack of variation.
Model 2: Facility Discharge
Approximately 38% of surgeries had a facility discharge (Table 2). Among the SES variables, patients that were ethnic minorities (OR 4.43, P<0.001), lived alone (OR 2.62, P<0.001), and weren’t working (OR 1.78, P=0.031) all independently predicted facility discharge. Among the adjusters (Table 4), surgeon, women, those with a Charlson score ≥2, year of surgery, and morbid obesity all predicted facility discharge. As expected, bilateral surgeries strongly predicted facility discharge, even after adjustment for other variables (OR 21.43, P<0.001). Older age, worse preoperative PCS, and worse preoperative MCS all had high dose-response predictive value for facility discharges. As with LOS, alcohol use was protective against facility discharge (OR 0.60, P<0.001). Without the 5 SES variables, the discharge predictive model had a validated and bootstrapped AUC of 0.86 (CI 0.84–0.87), indicating an excellent predictive value; adding the 5 SES variables only slightly improved the AUC to 0.87 (CI 0.85–0.88), which retains the excellent categorization. The difference between the AUCs was statistically significant (P=0.002).
Table 4.
Multivariate logistic model for whether separate SES variables individually predict facility discharge (37.9%) among primary TKAs (n=2,196)a
| With SES Variables | No SES Variables | |||||
|---|---|---|---|---|---|---|
| Variable | OR | 95% CI | P-Value | OR | 95% CI | P-Value |
| Race (ref = Non-Hispanic White) | ||||||
| Other | 4.43 | 2.03–9.64 | <0.001 | -- | -- | -- |
| Living Alone (ref = No) | ||||||
| Yes | 2.62 | 1.62–4.23 | <0.001 | -- | -- | -- |
| Missing | 0.83 | 0.44–1.58 | 0.577 | -- | -- | -- |
| Education (ref = Postgrad) | ||||||
| College | 0.92 | 0.50–1.67 | 0.778 | -- | -- | -- |
| Some College | 1.19 | 0.67–2.13 | 0.555 | -- | -- | -- |
| High School or Less | 1.01 | 0.56–1.82 | 0.965 | -- | -- | -- |
| Missing | 1.08 | 0.41–2.81 | 0.878 | -- | -- | -- |
| Employment (ref = Working/Student) | ||||||
| Retired, Not for Health | 1.25 | 0.75–2.06 | 0.390 | -- | -- | -- |
| Not Working | 1.78 | 1.05–3.01 | 0.031 | -- | -- | -- |
| Missing | 1.12 | 0.43–2.91 | 0.823 | -- | -- | -- |
| Household Annual Income (ref = $75,000+) | ||||||
| $50–75,000 | 1.06 | 0.50–2.28 | 0.875 | -- | -- | -- |
| $35–50,000 | 0.76 | 0.34–1.68 | 0.498 | -- | -- | -- |
| <$35,000 | 0.94 | 0.46–1.92 | 0.864 | -- | -- | -- |
| Refused to Answer | 0.58 | 0.29–1.15 | 0.117 | -- | -- | -- |
| Missing | 1.22 | 0.58–2.54 | 0.596 | -- | -- | -- |
| Surgeon (ref = Surgeon 1) | ||||||
| Surgeon 2 | 1.69 | 1.13–2.53 | 0.011 | 1.66 | 1.12–2.46 | 0.012 |
| Surgeon 3 | 0.48 | 0.15–1.55 | 0.220 | 0.51 | 0.16–1.57 | 0.239 |
| Surgeon 4 | 0.93 | 0.19–4.48 | 0.931 | 0.82 | 0.17–3.99 | 0.809 |
| Surgeon 5 | 1.52 | 0.69–3.31 | 0.295 | 1.72 | 0.79–3.76 | 0.176 |
| Surgeon 6 | 0.83 | 0.56–1.24 | 0.367 | 0.83 | 0.56–1.25 | 0.375 |
| Surgeon 7 | 1.06 | 0.74–1.51 | 0.752 | 1.10 | 0.78–1.55 | 0.598 |
| Surgeon 8 | 1.29 | 0.76–2.20 | 0.350 | 1.31 | 0.77–2.23 | 0.321 |
| Surgeon 9 | 0.45 | 0.19–1.08 | 0.073 | 0.47 | 0.19–1.15 | 0.099 |
| Surgeon 10 | 1.10 | 0.71–1.69 | 0.682 | 1.09 | 0.71–1.66 | 0.706 |
| Age Group (ref = <55) | ||||||
| 55–59 | 1.36 | 0.83–2.23 | 0.224 | 1.30 | 0.80–2.13 | 0.289 |
| 60–64 | 2.41 | 1.50–3.88 | <0.001 | 2.33 | 1.45–3.74 | <0.001 |
| 65–69 | 4.42 | 2.75–7.10 | <0.001 | 4.20 | 2.63–6.71 | <0.001 |
| 70–74 | 7.49 | 4.56–12.31 | <0.001 | 7.14 | 4.38–11.65 | <0.001 |
| 75–79 | 16.65 | 9.55–29.03 | <0.001 | 15.66 | 9.12–26.91 | <0.001 |
| 80+ | 20.38 | 11.22–37.02 | <0.001 | 18.75 | 10.44–33.67 | <0.001 |
| Sex (ref = Male) | ||||||
| Female | 2.00 | 1.56–2.57 | <0.001 | 2.07 | 1.62–2.66 | <0.001 |
| Preoperative Alcohol Use (ref = No) | ||||||
| Yes | 0.60 | 0.47–0.78 | <0.001 | 0.60 | 0.47–0.76 | <0.001 |
| Preoperative Tobacco Use (ref = Never) | ||||||
| Quit | 1.03 | 0.80–1.33 | 0.792 | 1.06 | 0.83–1.36 | 0.620 |
| Yes | 1.22 | 0.71–2.11 | 0.477 | 1.29 | 0.76–2.19 | 0.353 |
| Charlson Score (ref = 0) | ||||||
| 1 | 1.23 | 0.92–1.66 | 0.162 | 1.23 | 0.92–1.64 | 0.163 |
| 2+ | 1.55 | 1.14–2.10 | 0.005 | 1.53 | 1.13–2.06 | 0.005 |
| VR-12 PCS Preoperative (ref = 50+) | ||||||
| 40–49.99 | 2.38 | 1.33–4.23 | 0.003 | 2.53 | 1.44–4.44 | 0.001 |
| 30–39.99 | 2.53 | 1.47–4.37 | 0.001 | 2.74 | 1.60–4.68 | <0.001 |
| 20–29.99 | 3.43 | 2.02–5.82 | <0.001 | 3.69 | 2.19–6.22 | <0.001 |
| <20 | 4.91 | 2.66–9.05 | <0.001 | 5.25 | 2.89–9.52 | <0.001 |
| VR-12 MCS Preoperative (ref = 60+) | ||||||
| 50–59.99 | 1.64 | 1.20–2.24 | 0.002 | 1.64 | 1.21–2.21 | 0.001 |
| 40–49.99 | 2.53 | 1.75–3.66 | <0.001 | 2.59 | 1.81–3.73 | <0.001 |
| 30–39.99 | 2.91 | 1.89–4.47 | <0.001 | 2.94 | 1.92–4.49 | <0.001 |
| <30 | 2.29 | 1.32–3.99 | 0.003 | 2.26 | 1.32–3.85 | 0.003 |
| Year (ref = 2011) | ||||||
| 2012 | 1.38 | 0.89–2.14 | 0.145 | 1.43 | 0.94–2.19 | 0.098 |
| 2013 | 1.76 | 1.13–2.75 | 0.013 | 1.84 | 1.21–2.81 | 0.005 |
| 2014 | 1.40 | 0.86–2.27 | 0.171 | 1.48 | 0.93–2.34 | 0.098 |
| 2015 | 1.16 | 0.68–1.97 | 0.588 | 1.13 | 0.68–1.89 | 0.640 |
| 2016 | 1.70 | 0.74–3.93 | 0.213 | 1.67 | 0.75–3.73 | 0.208 |
| Bilateral TKA (ref = No) | ||||||
| Yes | 21.43 | 15.72–29.23 | <0.001 | 18.99 | 14.04–25.69 | <0.001 |
| BMI Preoperative (ref = Normal, <25) | ||||||
| Overweight, 25–29.99 | 0.92 | 0.62–1.36 | 0.661 | 0.93 | 0.63–1.36 | 0.694 |
| Obese, 30–34.99 | 1.27 | 0.85–1.91 | 0.245 | 1.30 | 0.88–1.93 | 0.193 |
| Severe Obese, 35–39.99 | 1.27 | 0.82–1.98 | 0.290 | 1.27 | 0.82–1.96 | 0.276 |
| Morbid Obese, 40+ | 3.89 | 2.44–6.20 | <0.001 | 3.81 | 2.40–6.03 | <0.001 |
| 10-fold cross-validation AUC | ||||||
| 0.87 | 0.85 – 0.88 | 0.86 | 0.84 – 0.87 | |||
Clustered on 2,059 patients and using robust standard errors to account for observational data
2 in-hospital deaths occurred and were not discharged.
Model 3: Clinically Significant PCS Improvement
Approximately 63% of surgeries resulted in clinically significant PCS improvement of at least 5 points30 among patients with both preoperative and postoperative PCS. Only 1 SES was associated with PCS: patients who were not working predicted against significant improvement (OR 0.55, P=0.031). Among the adjusting variables (Table 5), tobacco users and those with a Charlson score of 1 (but not 2+) predicted against significant improvement. Bilateral surgeries predicted improvement. Worse PCS and longer post-operative follow-up were associated with much greater odds of improvement in a dose-response fashion. Worse MCS predicted against improvement, also in a dose-response manner. Without the 5 SES variables, the PCS predictive model had a validated and bootstrapped AUC of 0.80 (CI 0.78–0.82), indicating excellent predictive value; adding the 5 SES variables only slightly improved the AUC to 0.81 (CI 0.79–0.83), which retains the excellent categorization. The difference between the AUCs was statistically significant (P=0.019).
Table 5.
Multivariate logistic model for whether separate SES variables individually predict clinically significant PCS improvement (5 points yes/no, 63% clinically significant improvement) among primary TKAs (n=1,723)
| With SES Variables | No SES Variables | |||||
|---|---|---|---|---|---|---|
| Variable | OR | 95% CI | P-Value | OR | 95% CI | P-Value |
| Race (ref = Non-Hispanic White) | ||||||
| Other | 0.78 | 0.34–1.79 | 0.556 | -- | -- | -- |
| Living Alone (ref = No) | ||||||
| Yes | 1.18 | 0.69–2.03 | 0.548 | -- | -- | -- |
| Missing | 1.36 | 0.66–2.77 | 0.405 | -- | -- | -- |
| Education (ref = Postgrad) | ||||||
| College | 0.89 | 0.49–1.62 | 0.698 | -- | -- | -- |
| Some College | 1.06 | 0.59–1.90 | 0.852 | -- | -- | -- |
| High School or Less | 1.43 | 0.76–2.67 | 0.266 | -- | -- | -- |
| Missing | 2.26 | 0.83–6.16 | 0.111 | -- | -- | -- |
| Employment (ref = Working/Student) | ||||||
| Retired, Not for Health | 0.67 | 0.39–1.14 | 0.136 | -- | -- | -- |
| Not Working | 0.55 | 0.32–0.95 | 0.031 | -- | -- | -- |
| Missing | 0.55 | 0.22–1.40 | 0.208 | -- | -- | -- |
| Household Annual Income (ref = $75,000+) | ||||||
| $50–75,000 | 0.74 | 0.33–1.67 | 0.469 | -- | -- | -- |
| $35–50,000 | 1.83 | 0.72–4.63 | 0.203 | -- | -- | -- |
| <$35,000 | 1.32 | 0.58–3.00 | 0.506 | -- | -- | -- |
| Refused to Answer | 0.98 | 0.49–1.98 | 0.961 | -- | -- | -- |
| Missing | 0.53 | 0.24–1.16 | 0.110 | -- | -- | -- |
| Surgeon (ref = Surgeon 1) | ||||||
| Surgeon 2 | 0.94 | 0.62–1.42 | 0.755 | 0.91 | 0.61–1.36 | 0.649 |
| Surgeon 3 * | -- | -- | -- | -- | -- | -- |
| Surgeon 4 | 0.68 | 0.24–1.89 | 0.457 | 0.66 | 0.23–1.86 | 0.431 |
| Surgeon 5 | 0.89 | 0.61–1.28 | 0.530 | 0.89 | 0.62–1.27 | 0.512 |
| Surgeon 6 | 0.92 | 0.65–1.31 | 0.645 | 0.91 | 0.64–1.28 | 0.574 |
| Surgeon 7 | 0.65 | 0.34–1.26 | 0.202 | 0.62 | 0.32–1.19 | 0.149 |
| Surgeon 8 | 1.49 | 0.64–3.48 | 0.353 | 1.21 | 0.53–2.75 | 0.646 |
| Surgeon 9 | 0.92 | 0.60–1.41 | 0.695 | 0.89 | 0.58–1.35 | 0.574 |
| Age Group (ref = <55) | ||||||
| 55–59 | 0.82 | 0.52–1.28 | 0.375 | 0.80 | 0.51–1.24 | 0.314 |
| 60–64 | 1.24 | 0.81–1.91 | 0.315 | 1.18 | 0.78–1.79 | 0.438 |
| 65–69 | 1.00 | 0.65–1.55 | 0.999 | 0.93 | 0.61–1.42 | 0.747 |
| 70–74 | 0.94 | 0.58–1.53 | 0.808 | 0.88 | 0.55–1.40 | 0.586 |
| 75–79 | 0.79 | 0.45–1.37 | 0.398 | 0.72 | 0.42–1.22 | 0.221 |
| 80+ | 0.63 | 0.36–1.10 | 0.105 | 0.61 | 0.35–1.06 | 0.080 |
| Sex (ref = Male) | ||||||
| Female | 1.07 | 0.83–1.37 | 0.609 | 1.05 | 0.82–1.34 | 0.702 |
| Preoperative Alcohol Use (ref = No) | ||||||
| Yes | 1.06 | 0.82–1.37 | 0.663 | 1.01 | 0.79–1.31 | 0.912 |
| Preoperative Tobacco Use (ref = Never) | ||||||
| Quit | 0.83 | 0.64–1.07 | 0.154 | 0.83 | 0.64–1.07 | 0.151 |
| Yes | 0.59 | 0.37–0.96 | 0.033 | 0.58 | 0.36–0.92 | 0.020 |
| Charlson Score (ref = 0) | ||||||
| 1 | 0.71 | 0.53–0.95 | 0.023 | 0.71 | 0.53–0.95 | 0.019 |
| 2+ | 0.85 | 0.62–1.16 | 0.306 | 0.83 | 0.61–1.13 | 0.235 |
| VR-12 PCS Preoperative (ref = 50+) | ||||||
| 40–49.99 | 6.31 | 3.22–12.36 | <0.001 | 6.23 | 3.22–12.07 | <0.001 |
| 30–39.99 | 25.69 | 13.25–49.81 | <0.001 | 24.80 | 12.89–47.74 | <0.001 |
| 20–29.99 | 38.79 | 19.95–75.39 | <0.001 | 37.25 | 19.33–71.79 | <0.001 |
| <20 | 57.48 | 26.97–122.48 | <0.001 | 54.09 | 25.66–114.05 | <0.001 |
| VR-12 MCS Preoperative (ref = 60+) | ||||||
| 50–59.99 | 0.99 | 0.73–1.34 | 0.948 | 0.97 | 0.72–1.31 | 0.836 |
| 40–49.99 | 0.51 | 0.35–0.73 | <0.001 | 0.50 | 0.35–0.70 | <0.001 |
| 30–39.99 | 0.42 | 0.27–0.65 | <0.001 | 0.42 | 0.28–0.65 | <0.001 |
| <30 | 0.35 | 0.20–0.59 | <0.001 | 0.35 | 0.21–0.60 | <0.001 |
| VR-12 PCS Post-Op Time Period (ref = 0 – 45 Days) | ||||||
| 46 – 299 Days | 4.51 | 3.10–6.55 | <0.001 | 4.43 | 3.06–6.39 | <0.001 |
| 300 – 420 Days | 7.15 | 4.73–10.79 | <0.001 | 6.86 | 4.55–10.32 | <0.001 |
| 421 + Days | 7.09 | 4.34–11.60 | <0.001 | 6.90 | 4.23–11.27 | <0.001 |
| Year (ref = 2011) | ||||||
| 2012 | 0.96 | 0.60–1.53 | 0.863 | 0.96 | 0.61–1.51 | 0.850 |
| 2013 | 1.00 | 0.62–1.60 | 0.989 | 1.00 | 0.64–1.58 | 0.983 |
| 2014 | 0.68 | 0.40–1.14 | 0.143 | 0.69 | 0.41–1.14 | 0.145 |
| 2015 | 0.86 | 0.46–1.61 | 0.643 | 0.88 | 0.48–1.61 | 0.678 |
| Bilateral TKA (ref = No) | ||||||
| Yes | 1.77 | 1.30–2.42 | <0.001 | 1.75 | 1.29–2.37 | <0.001 |
| BMI Preoperative (ref = Normal, <25) | ||||||
| Overweight, 25–29.99 | 0.96 | 0.65–1.42 | 0.854 | 0.92 | 0.63–1.36 | 0.689 |
| Obese, 30–34.99 | 0.70 | 0.46–1.06 | 0.092 | 0.66 | 0.44–1.01 | 0.053 |
| Severe Obese, 35–39.99 | 0.76 | 0.48–1.21 | 0.252 | 0.73 | 0.46–1.15 | 0.171 |
| Morbid Obese, 40+ | 0.71 | 0.44–1.14 | 0.155 | 0.68 | 0.42–1.08 | 0.102 |
| 10-fold cross-validation AUC | ||||||
| 0.81 | 0.79–0.83 | 0.80 | 0.78–0.82 | |||
Clustered on 1,723 patients and using robust standard errors to account for observational data
Surgeon 3 performed 2 surgeries with pre-operative and post-operative VR-12 scores and was excluded from analysis.
Among patients that completed VR-12s at multiple post-operative time periods, the order of priority for this model was 300 – 420 days, more than 421 days, 46 – 299 days, and 0 – 45 days post-operative.
Overall SES Variables
At least 1 SES variable predicted each outcome after adjustment for several factors, including other SES variables. Race/ethnicity and employment status were significant in two models while living alone and household income were significant in one model. Education level did not predict any outcome. Patients missing a SES value did not vary statistically from other patients for any SES variable or TKA outcome (Tables 3–5). Additionally, the Charlson Comorbidity Index, preoperative PCS and MCS, as well as bilateral surgeries, were all significantly predictive across all outcomes (LOS, discharge, and PCS improvement).
Overall Model Predictive Values
The LOS model was of acceptable predictive value, while the facility discharge and clinically significant PCS improvement models were of excellent predictive value. Despite SES variables statistically improving the predictive value of each model, the inclusion of the 5 SES variables did not change the category of predictive value for any model, and improved the models’ value by either 0.01 or 0.02 on a 0.01 – 0.99 range. There were no differences in the values of cross-validated and bootstrapped AUC values across all models.
Sub-Analysis
In a sub-analysis (data not shown) that ran the same models but restricted to only patients that that had answered at least 1 SES variable, and typically answered all of them (n = 782, 804, and 631 respectively for each model), the results were generally similar with correspondingly lower statistical power. There were no differences in age, gender, Charlson, surgeon, or preoperative PCS/MCS between those patients who did and did not answer an SES question. The AUC difference for these smaller models, with and without SES variables for LOS, discharge, and PCS improvement was 0.04, 0.01, and 0.02, respectively.
Discussion
Among our 2,198 primary TKAs at a rural tertiary medical center, 22% of surgeries resulted in LOS over 3 days, 38% were sent to a facility after discharge, and 63% underwent clinically significant improvement in self-reported physical function. Our percentage of longer LOS compares favorably to a national sample of 25%, although our facility discharge rate is higher than a national total joint arthroplasty sample of 30%; when excluding bilateral surgeries, our facility discharge rate was 26%.26,38 Along with our results, other studies have noted that despite TKA’s cost-effectiveness,1,2 many patients may not achieve significant improvement in physical function as measured by PRO change score.15,39 In our work, 37% of surgeries did not result in clinically significant improvement in physical function. Future research should explore whether patient satisfaction has any relationship with their reported physical function improvement.
It is notable that all three models had at least “acceptable” predictability and two had “excellent” predictability using rigorous statistical techniques, with or without the SES variables. Additionally, none of our variables were clinical in nature, or were even specific to the knee: theoretically, all of our data could have been obtained using surveys without a physical visit. Other orthopaedic studies have noted the power of predictive models. Keeney et al in 2013 noted a cross-validated AUC of 0.89 for predicting spine surgery after non-traumatic occupational back injuries using only 3 variables, none of which were SES or demographic in nature; adding 17 more variables, all of which were significant bivariately, only minutely improved the cross-validated AUC to 0.93.23
In our “acceptable” LOS > 3 days model, ethnic minorities and low household income predicted longer LOS, while surgeon, age ≥75, Charlson score over 0, lower PCS, lower MCS, bilateral TKA, morbidly obese, and year of surgery also predicted longer LOS; alcohol use predicted lower LOS. These SES findings are similar to a national sample, which also reported that race and low incomes were associated with longer LOS, though that study did not have any other SES variables available.26 Another study noted the association between income and LOS, although that study used census geographic income, didn’t adjust for other SES factors, and did not provide predictive models.20 We could not locate another preoperative predictive model for longer TKA LOS among American hospitals.
Among the SES variables in our “excellent” predictive model for facility discharge, values for ethnic minority status, living alone, and not working individually predicted facility discharge. These models achieved very high predictive values of 0.87 and 0.86. Other studies have noted that ethnic minority status was associated with TKA discharge destination.40,41 We could not locate another study that included living alone or unemployment as a predictor of TKA facility discharge, though Rissman 2016 noted that missing a status of living with others as a limitation, and Courtney 2016 noted a relationship between census-based income and discharge.12,20
In our “excellent” predictive model for clinically significant PCS improvement, the only significant SES variable was “not working”, which predicted lower odds of improvement. Goodman 2016 also noted an association between race/ethnicity and education with PROs (2-year WOMAC), but that the association was clinically small and did not meet their threshold.19 Our AUC model values of 0.81 and 0.80 improve on another medical center’s model for PCS improvement of 5 points at 1 year, which had a multivariate predictive AUC value of 0.71 with fewer variables.15 That model included SF-12 PCS and MCS, gender, age, and race. In a sub-analysis, we matched their variables to our data with only 1-year PCS follow-up measures and achieved an AUC of 0.77 (95% CI 0.73–0.81) (data not shown). With only 5 variables, none of which required a clinical visit or were specific to the knee, we obtained an AUC score within 0.04 of our larger multivariate model with 17 variables (Table 5).
These combined findings, combined with earlier papers,15,23 give increasing credence that there are diminishing returns for adding more variables to predictive surgical models and that specific variables, especially preoperative physical function, are more important to collect than others.
Our study contains many limitations. Our missingness rates are high among all of our SES variables. It is possible that a higher capture rate may change our results, although our predictive models already present strong predictive ability. However, our patients who did not answer SES questions did not vary from those that did, and a smaller sub-model of surgeries with full SES capture had similar findings. Although our results were internally cross-validated and bootstrapped, our patient population consists of one large rural academic medical center and does not present substantial racial and ethnic diversity, although it is representative of the region and 88% of TKAs in the Medicare population continue to occur in the white population.12,41 However, our population did present a large range of the other SES variables. Lastly, SES variables may remain helpful for determining systematic outcomes or outreach concerns between different cohorts of patients across different institutions or in risk adjustment for other outcomes not documented here. However, we note that we have achieved excellent and acceptable predictive models without the inclusion of any SES variables for our outcomes of interest. Our study also contains several strengths: we have identified strong prospective predictive models for longer LOS, facility discharge, and clinically significant physical function improvement using only variables that may be determined by patients while accounting for numerous separate SES indicators.
Conclusions
We included several SES variables in multivariate predictive models for TKA outcomes in a rural area. At least 1 SES variable contributed individually to each model and the overall models statistically improved with their inclusion. However, the predictive improvement of the SES models was minor and one must consider whether the improved predictive power of AUC 0.01 or 0.02 justifies the additional efforts to collect these variables. As an institution, we will continue to collect these variables (race/ethnicity, living situation, education, employment, and household income) as they are already embedded in our patient pathway. However, other institutions may want to focus on other variables that offer more benefit for predictive modeling in the current joint replacement climate, particularly patient-reported measures like physical and mental function, prior to incorporating SES variables into their clinic workflow.
Acknowledgments
Funding sources
This work was supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) (P60-AR048094 and P60-AR062799).
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Daigle ME, Weinstein AM, Katz JN, Losina E. The cost-effectiveness of total joint arthroplasty: a systematic review of published literature. Best practice & research. Clin Rheum. 2012;26(5):649–58. doi: 10.1016/j.berh.2012.07.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Jenkins PJ, Clement ND, Hamilton DF, Gaston P, Patton JT, Howie CR. Predicting the cost-effectiveness of total hip and knee replacement: a health economic analysis. Bone Joint J. 2013;95-B(1):115–21. doi: 10.1302/0301-620X.95B1.29835. [DOI] [PubMed] [Google Scholar]
- 3.Weiss AJ, Elixhauser A, Andrews RM. HCUP Statistical Brief #170. Agency for Healthcare Research and Quality; Rockville, MD: Feb, 2014. Characteristics of operating room procedures in U.S. hospitals, 2011. [PubMed] [Google Scholar]
- 4.Healy WL, Rana AJ, Iorio R. Hospital economics of primary total knee arthroplasty at a teaching hospital. Clin Ortho Rel Res. 2011;469(1):87–94. doi: 10.1007/s11999-010-1486-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kurtz SM, Ong KL, Lau E, Bozic KJ. Impact of the economic downturn on total joint replacement demand in the United States: updated projections to 2021. J Bone Joint Surg Am. 2014;96(8):624–30. doi: 10.2106/JBJS.M.00285. [DOI] [PubMed] [Google Scholar]
- 6.Bosco JA, Alvarado CM, Slover JD, Iorio R, Hutzler LH. Decreasing total joint implant costs and physician specific cost variation through negotiation. J Arthroplasty. 2014;29(4):678–80. doi: 10.1016/j.arth.2013.09.016. [DOI] [PubMed] [Google Scholar]
- 7.Tomek IM, Sabel AL, Froimson MI, Muschler G, Jevsevar DS, Koenig KM, Lewallen DG, Naessens JM, Savitz LA, Westrich JL, Weeks WB, Weinstein JN. A collaborative of leading health systems finds wide variations in total knee replacement delivery and takes steps to improve value. Health Aff. 2012;31(6):1329–38. doi: 10.1377/hlthaff.2011.0935. [DOI] [PubMed] [Google Scholar]
- 8.Casalino LP, Gans D, Weber R, Cea M, Tuchovsky A, Bishop TF, Miranda Y, Frankel BA, Ziehler KB, Wong MM, Evenson TB. US physician practices spend more than $15. 4 billion annually to report quality measures. Health Affairs. 2016;35(3):401–6. doi: 10.1377/hlthaff.2015.1258. [DOI] [PubMed] [Google Scholar]
- 9.Forrest G, Fuchs M, Gutierrez A, Girardy J. Factors affecting length of stay and need for rehabilitation after hip and knee arthroplasty. J Arthroplasty. 1998;13(2):186–90. doi: 10.1016/s0883-5403(98)90097-3. [DOI] [PubMed] [Google Scholar]
- 10.Ong PH, Pua YH. A prediction model for length of stay after total and unicompartmental knee replacement. Bone Joint J. 2013;95-B(11):1490–6. doi: 10.1302/0301-620X.95B11.31193. [DOI] [PubMed] [Google Scholar]
- 11.Weaver F, Hynes D, Hopkinson W, Wixson R, Khuri S, Daley J, Henderson WG. Preoperative risks and outcomes of hip and knee arthroplasty in the Veterans Health Administration. J Arthroplasty. 2003;18(6):693–708. doi: 10.1016/s0883-5403(03)00259-6. [DOI] [PubMed] [Google Scholar]
- 12.Rissman CM, Keeney BJ, Ercolano EM, Koenig KM. Predictors of facility discharge, range of motion, and patient-reported physical function improvement after primary total knee arthroplasty: A prospective cohort analysis. J Arthroplasty. 2016;31(1):36–41. doi: 10.1016/j.arth.2015.09.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Smith ID, Elton R, Ballantyne JA, Brenkel IJ. Pre-operative predictors of the length of hospital stay in total knee replacement. J Bone Joint Surg Br. 2008;90(11):1435–40. doi: 10.1302/0301-620X.90B11.20687. [DOI] [PubMed] [Google Scholar]
- 14.Raut S, Mertes SC, Muniz-Terrera G, Khanduja V. Factors associated with prolonged length of stay following a total knee replacement in patients aged over 75. Int Orthop. 2012;36(8):1601–8. doi: 10.1007/s00264-012-1538-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Berliner JL, Brodke DJ, Chan V, SooHoo NF, Bozic KJ. Can preoperative patient-reported outcome measures be used to predict meaningful improvement in function after TKA? Clin Orthop Relat Res. 2016 Mar 8; doi: 10.1007/s11999-016-4770-y. Epub ahead of print. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.D’Apuzzo MR, Novicoff WM, Browne JA. The John Insall Award: Morbid obesity independently impacts complications, mortality, and resource use after TKA. Clin Orthop Relat Res. 2015;473(1):57–63. doi: 10.1007/s11999-014-3668-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Barsoum WK, Murray TG, Klika AK, Green K, Miniaci SL, Wells BJ, Kattan MW. Predicting patient discharge disposition after total joint arthroplasty in the United States. J Arthroplasty. 2010;25(6):885–92. doi: 10.1016/j.arth.2009.06.022. [DOI] [PubMed] [Google Scholar]
- 18.Kremers HM, Kremers WK, Berry DJ, Lewallen DG. Social and behavioral factors in total knee and hip arthroplasty. J Arthroplasty. 2015;30:1852–4. doi: 10.1016/j.arth.2015.04.032. [DOI] [PubMed] [Google Scholar]
- 19.Goodman SM, Mandl LA, Parks ML, Zhang M, McHugh KR, Lee Y-Y, Nguyen JT, Russell LA, Bogardus MH, Figgie MP, Bass AR. Disparities in TKA outcomes: Census tract data show interactions between race and poverty. Clin Orthop Relat Res. 2016;474:1986–95. doi: 10.1007/s11999-016-4919-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Courtney PM, Huddleston JI, Iorio R, Markel DC. Socioeconomic risk adjustment models for reimbursement are necessary in primary total joint arthroplasty. In-press. J Arthroplasty. 2016 doi: 10.1016/j.arth.2016.06.050. [DOI] [PubMed] [Google Scholar]
- 21.Zhang W, Lyman S, Boutin-Foster C, Parks ML, Pan T-J, Lan A, Ma Y. Racial and ethnic disparities in utilization rate, hospital volume, and perioperative outcomes after total knee arthroplasty. J Bone Joint Surg Am. 2016;98:1243–52. doi: 10.2106/JBJS.15.01009. [DOI] [PubMed] [Google Scholar]
- 22.Katz JN. Editorial: Persistence of racial and ethnic differences in utilization and adverse outcomes of total joint replacement. J Bone Joint Surg Am. 2016;98:1241–2. doi: 10.2106/JBJS.16.00623. [DOI] [PubMed] [Google Scholar]
- 23.Keeney BJ, Fulton-Kehoe D, Turner JA, Wickizer TM, Chan KCG, Franklin GM. Early predictors of lumbar spine surgery after occupational back injury: Results from a prospective study of workers in Washington State. Spine. 2013;38(11):953–64. doi: 10.1097/BRS.0b013e3182814ed5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.American Joint Replacement Registry. What Data Do We Collect? www.ajrr.net/enroll-with-us/data-elements-we-collect. Last checked on 3/22/2016.
- 25.Franklin PD, Lewallen D, Bozic K, Hallstrom B, Jiranek W, Ayers DC. Implementation of patient-reported outcome measures in U.S. total joint replacement registries: Rationale, status, and plans. J Bone Joint Surg Am. 2014;96(Supp 1):104–9. doi: 10.2106/JBJS.N.00328. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.el Bitar YF, Illingworth KD, Scaife SL, Horberg JV, Saleh KJ. Hospital length of stay following primary total knee arthroplasty: Data from the Nationwide Inpatient Sample Database. J Arthroplasty. 2015;30:1710–5. doi: 10.1016/j.arth.2015.05.003. [DOI] [PubMed] [Google Scholar]
- 27.Sørensen LT. Wound healing and infection in surgery: The clinical impact of smoking and smoking cessation: A systematic review and meta-analysis. Arch Surg. 2012;147(4):373–83. doi: 10.1001/archsurg.2012.5. [DOI] [PubMed] [Google Scholar]
- 28.Kazis LE, Selim A, Rogers W, Ren XS, Lee A, Miller DR. Dissemination of methods and results from the veterans health study: Final comments and implications for future monitoring strategies within and outside the veterans healthcare system. J Ambul Care Manage. 2006;29:310–9. doi: 10.1097/00004479-200610000-00007. [DOI] [PubMed] [Google Scholar]
- 29.Selim AJ, Rogers W, Fleishman JA, Qian SX, Fincke BG, Rothendler JA, Kazis LE. Updated U.S. population standard for the Veterans RAND 12-Item Health Survey (VR-12) Qual Life Res. 2009;18:43–52. doi: 10.1007/s11136-008-9418-2. [DOI] [PubMed] [Google Scholar]
- 30.SooHoo NF, Li Z, Chenok KE, Bozic KJ. Responsiveness of patient reported outcome measures in total joint arthroplasty patients. J Arthroplasty. 2015;30:176–91. doi: 10.1016/j.arth.2014.09.026. [DOI] [PubMed] [Google Scholar]
- 31.Khatib Y, Madan A, Naylor JM, Harris IA. Do psychological factors predict poor outcome in patients undergoing TKA? A systematic review. Clin Orthop Relat Res. 2015;473(8):2630–8. doi: 10.1007/s11999-015-4234-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Halawi MJ, Vovos TJ, Green CL, Wellman SS, Attarian DE, Bolognesi MP. Preoperative predictors of extended hospital length of stay following total knee arthroplasty. J Arthroplasty. 2015;30(3):361–4. doi: 10.1016/j.arth.2014.10.025. [DOI] [PubMed] [Google Scholar]
- 33.Ayers DC, Franklin PD. Joint replacement registries in the United States: A new paradigm. J Bone Joint Surg Am. 2014;96(18):1567–9. doi: 10.2106/JBJS.N.00641. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Stata Statistical Software. Version 12MP. College Station, TX: StataCorp LP; 2012. [Google Scholar]
- 35.Steyerberg EW, Harrell FE, Jr, Borsboom GJJM, et al. Internal validation of predictive models: Efficiency of some procedures for logistic regression analyses. J Clin Epidemiol. 2001;54:774–81. doi: 10.1016/s0895-4356(01)00341-9. [DOI] [PubMed] [Google Scholar]
- 36.Hosmer DW, Lemeshow S. Applied Logistic Regression. 2. New York, NY: John Wiley; 2000. [Google Scholar]
- 37.Janes H, Longton G, Pepe MS. Accommodating covariates in receiver operating characteristic analysis. Stata J. 2009;9(1):17–39. [PMC free article] [PubMed] [Google Scholar]
- 38.Keswani A, Tasi MC, Fields A, Lovey AJ, Moucha CS, Bozic KJ. Discharge destination after total joint arthroplasty: An analysis of postdischarge outcomes, placement risk factors, and recent trends. J Arthroplasty. 2016;31:1155–62. doi: 10.1016/j.arth.2015.11.044. [DOI] [PubMed] [Google Scholar]
- 39.Alzahrani K, Gandhi R, Debeer J, Petruccelli D, Mahomed N. Prevalence of clinically significant improvement following total knee replacement. J Rheumatol. 2011;38:753–9. doi: 10.3899/jrheum.100233. [DOI] [PubMed] [Google Scholar]
- 40.Schwarzkopf R, Ho J, Quinn JR, Snir N, Mukamel D. Factors influencing discharge destination after total knee arthroplasty: A database analysis. Geriatr Orthop Surg Rehabil. 2016;7(2):95–9. doi: 10.1177/2151458516645635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Singh JA, Lu X, Rosenthal GE, Ibrahim S, Cram P. Racial disparities in knee and hip total joint arthroplasty: An 18-year analysis of national Medicare data. Ann Rheum Dis. 2014;73(12):2107–15. doi: 10.1136/annrheumdis-2013-203494. [DOI] [PMC free article] [PubMed] [Google Scholar]