Readmission, Complication, and Disposition Calculators in Total Joint Arthroplasty: A Systemic Review

Abstract

Background:

Predictive tools are useful adjuncts in surgical planning. They help guide patient selection, candidacy for inpatient versus outpatient surgery, and discharge disposition as well as predict the probability of readmissions and complications after total joint arthroplasty (TJA). Surgeons may find it difficult due to significant variation among risk calculators to decide which tool is best suited for a specific patient for optimal decision-based care. Our aim is to perform a systematic review of the literature to determine the existing post-TJA readmission calculators and compare the specific elements that comprise their formula. Secondly, we intend to evaluate the pros and cons of each calculator.

Methods:

Using a PRISMA protocol, we conducted a systematic search through three major databases for publications addressing TJA risk stratification tools for readmission, discharge disposition and early complications. We excluded those manuscripts that were not comprehensive for hips and knees, did not list discharge, readmission or complication as the primary outcome, or were published outside the North America.

Results:

Ten publications met our criteria and were compared on their sourced data, variable types, and overall algorithm quality. Seven of these were generated with single institution data and three from large administrative datasets. Three tools determined readmission risk, five calculated discharge disposition, and two predicted early complications. Only 4 prediction tools were validated by external studies. Seven studies utilized preoperative data points in their risk equations while three utilized intraoperative or post-surgical data to delineate risk.

Conclusion:

The extensive variation among TJA risk calculators underscores the need for tools with more individualized stratification capabilities and verification. The transition to outpatient and same day discharge TJA may preclude or change the need for many of these calculators. Further studies are needed to develop more streamlined risk calculator tools that predict readmission and surgical complications.

Introduction:

Decision-aids are increasingly used in surgical subspecialties to increase patient knowledge of a condition, improve surgical outcomes and enhance patient-centered decision-making [1]. Stratification tools have been developed to help guide patient selection, determine candidacy for inpatient versus outpatient total joint arthroplasty (TJA), predict discharge disposition, forecast postoperative narcotic use, and calculate risks of complications, such as fracture and infection[2–11].

Preventing hospital readmissions during the episode of care is an indispensable part of a successful and profitable TJA practice operating under alternative payment models [12–14]. With the current tools available for risk stratification, confusion exists amongst physician teams when trying to select a particular calculator to best stratify their patient’s risk profile. Some studies are based on retrospective associations (odds ratios).Other calculators aim to prospectively calculate a cumulative risk for a given complication, like readmission, based on an individual patient’s demographic, psychosocial and physiologic characteristics [15]. Despite these risk stratification guides, there is a paucity of information regarding specific variables utilized in each calculator and their relative weight to determine a specific outcome.

It is our primary aim to perform a systematic review of the literature to determine the existing post-TJA readmission calculators and compare the specific elements that comprise each one’s weighted formula. Secondly, we intend to evaluate the pros and cons of each calculator based on its intended use and prior validation.

Materials & Methods:

Our study methods followed the initial criteria of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Protocols (PRISMA-P) [16]. To identify relevant published manuscripts regarding prediction tools or calculators related to the risk of hospital, discharge, complications and readmission, we conducted comprehensive searches utilizing the PubMed/Medline and Embase search engine database. Our initial search utilized MeSH terms of “arthroplasty” and “readmission” and an associated search adding “risk calculator.” We set limitations to only include publications written in English, formulated for or validated to patients in United States, and published after 2005 to account for modern arthroplasty techniques and protocols. To capture potential articles that were not found in the initial MeSH search, we ran a second search with the terms “arthroplasty” and “stratification” and “calculator.” Next, we conducted searches using the exploded MeSH term of “readmission” and “risk” and “complication” and the key words “model”, “predict,” and “discharge.” Finally, we reviewed the citations of potential articles as a back-wall effort to find articles missed from our prior search strategy.

Our search efforts discovered 922 unique potential publications. One senior author (JBS) screened and reviewed the abstracts to exclude those records not involving hip and knee arthroplasty, those published outside the United States, and those not listing readmission, discharge, or early complication as the primary outcome (Figure 1). Secondary exclusions after full-text review were then made for the remaining manuscripts to remove those that included revision or conversion arthroplasty, management of infection or fracture, not including both hip and knee replacement, and or not including predictive analysis or development a risk stratification tool or formula. After the rigorous exclusion process, 10 studies were left for in-depth review regarding risk stratification methods surrounding the TJA episode of care. Risk calculator publications were then compared for their unique input variables, patient populations and grouped based on their primary outcome.

Figure 1.

PRISMA 2009 flow diagram

Results:

Readmission Calculators

Five risk calculators were built for the purposes of assessing postoperative readmissions and complications. Two of these tools, from the American College of Surgeons (ACS) and the American Joint Replacement Registry (AJRR), based their models on larger administrative databases, both projecting complications in addition to readmission. Three tools were built from single institution data and focus on readmission probabilities. The Duke calculator is a 90-day readmission risk predictor including both intra- and post-operative variables. The OrthoCincy readmission tool determines risk probabilities for both 30 and 90-day readmissions and the Readmission Risk Assessment Tool (RRAT) focuses on modifiable risk factors in projecting 30-day readmission risks.

Total Joint Replacement Risk Calculator (AJRR)

In 2012, the AJRR published the Total Joint Replacement risk calculator developed by Ortho Apps (a collaboration of Massachusetts General Hospital, University of California at San Francisco, and Mayo Clinic). The intended purpose of the tool was for surgeons to input their patient’s demographics and over 30 relevant comorbidities to assess potential risks for TJR procedures (Table 3). Risks assessed included mortality within 90 days and periprosthetic joint infection within two years. Because the tool was created around data from the Center for Medicare & Medicaid Services (CMS) between 2007–2012, a patient’s risks could be graphed and compared to both a patient with similar demographics and no comorbidities or the Medicare population average [9,17].

Table 3.

Review of benefits and drawbacks of risk stratification guides to predict outcomes and readmissions in total joint arthroplasty

Risk Stratification Tool	Pros	Cons
ACS	Able to compare a patient’s risks to the average patient	Validation Study: Low predictability. Assessing risk for all surgeries; probabilities could vary from assessing risk for TJA alone
AJRR	Able to compare patient’s risks to both their counterparts with similar demographics minus comorbidities and to Medicare population at large	Validation Study: Poor discrimination. Includes 30 comorbidity and patient factor variables, which decreases practical use and confounds contribution of any single variable.
Duke	Single institution's data, which renders input variables more specific to arthroplasty population and not reliant on administrative coding	Includes intra- and post-operative variables, which limits its utility in trying to control for and address complications before surgery
OrthoCincy	Single institution's data, which renders input variables more specific to arthroplasty population and not reliant on administrative coding	Not a true calculator, but a retrospective study to see what factors associate more with readmission
RRAT	Gives providers data preoperatively that can be used to optimize and address modifiable factors beforehand	Limited data input variables considered for chronic medical conditions
Iowa	Queried ACS NSQIP databases of all TJAs from 2011–2013 (comprehensive population)	Not a true calculator, but a retrospective study to see what factors associate more with certain discharge destinations. Also does not consider role of modifiable risk factors
Cleveland	Provides input variables unique to discharge (i.e. location of bedrooms and # of floors in house)	Doesnť specifically focus on complication or readmission risk
RAPT	Accounts for surgeon suspicion on discharge to home or a facility, and then tests their predicted preferences based on patient variables. Validation Study: accuracy of the risk ranges determined by RAPT developers	This tool mainly quantifies the doctor's subjective preferences, as part of the methods had doctors independentlydecide who would they thought be sent home and who would be discharged to a facility. Then they based their low/high risk criteria ranges from the differences of six variables from questionnaires from those who were sent home and sent to facility. Uses data from 3 different patient populations (TJA, spinal fusions, and valve replacements) and could present with different results if TJA was exclusively studied.
PARS	Has validation study for episode of care costs in which higher score correlates with greater cost of care, but not readmission or initial length of stay.	Does not show data on tool’s effectiveness in clinical setting, as it requires multiple intra-operative variables for its application. Although their use of intraoperative variables may help predict critical care needs, it will not help in mitigating necessary risk factors prior to operations.
OARA	Validated in external studies as accurate screening tool for same-day discharging	Does not show data on tool’s effectiveness in clinical setting (i.e. readmission rates for determined low OARA scores). Internal validation studies showing high false negative rate show that patient’s with low scores are not being discharged (causing inefficiencies)

The validity of the AJRR risk calculator was challenged when applied it to more than 30,000 TJA cases within the Veterans Affairs Surgical Quality Improvement Program database. Unfortunately, the AJRR tool had poor discrimination in identifying patients at risk of death within 90 days of surgery with an area under the curve of 0.62 (95% CI, 0.60–0.64) [18]. There was substantial overlap in the model between those who did and did not die in 90 days. Due to concerns of generalizability of AJRR patients with the abundance of input variables and limited initial patient sample from the few participating sites, the first iteration of the calculator is no longer available for risk stratification use.

ACS risk calculator

The ASC National Surgical Quality Improvement Program (NSQIP) developed an open-access surgical risk calculator to predict the likelihood of certain post-operative complications across all surgical subspecialties. Their model is built using data from 4.3 million operations between 2013–2017 within the NSQIP database and has information regarding any events or complications within 30 days after surgery[19]. The tool also features a “Surgeon Adjustment of Risks” to allow for an increase in calculated risk by one or two standard deviations based on surgeon discretion of cumulative risk based on overall patient’s health.

The user can enter up to 22 possible risk factors related to a patient’s health, which allows for the tool to estimate risks of post-operative complications and mortality (Table 2). The calculator allows a user to choose from 1,500 procedures (listed by CPT codes) across all specialties. Based on procedure and patient inputs, the algorithm will output the patient’s percentage risk for complications including pneumonia, cardiac arrest, surgical site infections, DVTs, discharge to post-acute care facilities, readmission, and death among others. The calculator also outputs comparisons to the average patients’ risk of these complications plus a predicted length of the patient’s hospital stay. Significant post-operative complication prediction results include pneumonia, heart problems, surgical site infections, urinary tract infections, blood clotting, and kidney failure.

Table 2.

Specific comorbidities listed for each risk stratification tool.

Comorbidities	ACS	AJRR	Duke	OrthoCincy	RRAT	Cleveland	PARS	OARA
Diabetes	x	x	x	x	X	x
Vascular	Medicated HTN	HTN Peripheral Vasculardz	HTN (Complicated, controlled		Any (+ VTED)	HTN
		Cerebrovascular dz	Peripheral Vasculardx
Cardiac		Arrhythmia	Arrhythmia	Arrhythmia	Any	Arrhythmia	CAD	x
		CHF		Heart Failure			CHF
		Ischemic Heart
		Valvular disease
Substance Use		Drug Abuse	Drug Abuse	Drug Abuse
		Alcohol Abuse		Narcotic usage
Hematologic		Anemia						x
		Coagulopathy
		Hyper-cholesterolemia
Pulmonary	Dyspnea	Chronic Pulmonary				COPD	COPD	x
	COPD	Pulmonary circulation
	Ventilator usage
Cancer	Disseminated	Metastatic Tumor
		Malignancy
		Lymphoma
		Weight loss
Infection		HIV			nasal colonization			x
		UTI
Renal Disorder	Acute renal dx	Electrolyte dx						x
	Dialysis	Renal disease
GI		Chronic Liver dx	Liver disease					x
		Peptic ulcer
Medication				Med count *
Others	Chronic steroid	Dementia	Neurologic dx	10+ prior ED visits		Arthritis		GeneralMedical
		Hemiplegia, paraplegia						Endocrine
		Rheumatologic dx						Neurologic, Psych
		Hypothyroid

dx = disorder

^*prior to admission, HTN hypertension, VTED venous thrombotic embolic disease, CAD coronary artery disease, CHF congestive heart failure, COPD chronic obstructive lung disease

^†Iowa and RAPT do not include comorbidities

In an external validation study performed of 1066 primary elective TJAs (694 TKA, 372 THA) from 2009–2012 of Medicare’s database, the calculator yielded excellent predictive results for death, overall complications, and serious complication rates for knee and hip replacement surgery [20]. Each patient’s information was input into the calculator and its risk estimates were recorded and compared to the observed documented patient outcomes. In the overall risk analysis, ACS probabilities for cardiac complications (10.1 [3.1, 32.5], P < 0.001), pneumonia (170.5 [23.1, 1255.3], P < 0.001), facility discharge (1.1 [1.1, 1.1], P < 0.001), and any complication overall (OR with 95% CI: 1.2 [1, 1.3], P = 0.005)were significantly associated with their respective actual event occurrence. Although given a small c-statistic (0.62), ACS’s risk estimates of pneumonia demonstrated the strongest predictability of readmission within 30 days for the 6.4% of patients that were readmitted in the study (OR = 8.28 [3.1, 22.4], P < 0.05). The review determined predictability to be low across all complications[21]. Thus, while the ACS tool is strengthened by its large procedural database, the input variables are not suited to provide strong predictions for specific surgical complications of TJAs in the Medicare population.

Duke readmission calculator

A group from Duke University created a system to predict readmission within 90 days of a TJA procedure. The authors performed a query of the medical records for 5570 knee and 4585 hip patients excluding bilateral cases, revisions, and arthroplasty for fractures between 2013to 2018[3]. Univariable logistic regression was applied to examine for independent associations with readmissions. Seventeen particular variables were identified for a final multivariable logistic regression through backward stepwise elimination and parameter selection of potential variables in predicting readmission (Table 2). These seventeen variables were used to create a nomograph to measure a patient’s overall likelihood of 90-day readmission based on the weighted variable contribution. The nomogram was used with internal institutional data to automate estimations of probabilities of readmission with 95% confidence intervals based on the patient’s inputs. To date, no external validation studies have been done to test the generalizability of this group’s predictive nomograph.

OrthoCincy readmission tool

OrthoCincy, a multi-specialty group of Orthopaedic Surgeons in Greater Cincinnati (Edgewood, KY), developed a model to predict 30 and 90-day readmissions after TJA. To build their model, binary logistic regressions were used in a retrospective analysis of 5732 TJA patients from their own practice between 2013 to 2018[10]. Within this sample, 237 patients were readmitted within 30 days and 547 within 90 days. Analysis of the 30-day readmissions showed associations with age, BMI, gender, discharge disposition, prior ER visits, psychiatric diagnoses, and medication counts prior to admissions, dysrhythmia, and heart failure. Patients with higher preoperative medication counts showed a direct correlation to higher risk of readmission. 90-day readmissions had the same associations except for age and psychiatric diagnoses, but additionally included intravenous drug abuse, narcotic medications, and TJA within a year. Original discharge to a facility after index TJA demonstrated the greatest risk of readmission (r² 0.3645; SE 0.0999) while home-health utilization showed moderate risk, and self-care had the lowest risk (r².3124; SE 0.1009). From their analysis, they generated separate estimated regression equations for 30- and 90-day readmissions involving twelve different demographic, medical comorbidity and social factors and precision estimates of +/−5.1% for 30-day readmissions and +/− 7.4% for the 90-day model. These regression equations have yet to be tested by inputting data from patients outside of the developer’s practice. This risk calculator, however, includes a postoperative variable (discharge deposition) and cannot be used as a predictive tool before surgery.

Readmission Risk Assessment Tool

The RRATseeks to quantify total hip and knee arthroplasty patient’s modifiable risk factors for stratification of readmissions. The tool was developed at New York University’s Hospital for Joint Diseases using a nested case-control cohort of readmitted (n=207) and non-readmitted patients (n=234) with a random and an age-matched control cohort between 2008 to 2013[20]. Eight modifiable risk factors were selected to generate the RRAT. The patients were then scored from1–3 on each factor’s severity with the cumulative sum being their RRAT total. The three risk factors that were significantly associated with readmission regardless of age were diabetes, history of VTED, and smoking[20]. The odds given for each discrete score gave a ratio of patients who were readmitted versus those that were not readmitted with that given score. A given score’s odds ratio calculated the odds of that score divided by the odds of the previous score. Total scores greater than 3 were found to have higher odds of readmission (scores of 1 and 2 have odds of 0.52 and 0.92 respectively) and scores > 5 yielded significantly greater odds of readmission (OR 5.31 versus a score of 4, with an OR of 2.97). However the odds ratio stayed equal at 1.79 across all scores from 1 to 6. 45% of readmission cases were from surgical site infection, 32% were from medical complications, and 23% from other arthroplasty-related complications. The NYU group reviewed the effect of their Perioperative surgical home (POSH) in improving readmission rates based on stratification using the RRAT tool. They found that the POSH significantly lowered readmission rates in the patients with an RRAT score greater than 3 [22]. They recommend using the RRAT to help determine which patients may have modifiable risk factors that can be improved before surgery.

Discharge Calculators

Five risk calculators in our study aimed to project appropriate postoperative discharge dispositions for TJA patients. The Iowa tool is built from the NSQIP database and is meant to determine discharge disposition. The remaining four calculators were built from data at a single institution. The Cleveland Clinic calculator determines postop discharge destination. The Risk Assessment and Prediction Tool (RAPT) includes variables for the patient’s functional status and is meant to determine the need for inpatient rehabilitation. The Penn Arthroplasty Risk Score (PARS) calculates need for post-operative intensive care while The Outpatient Arthroplasty Risk Assessment (OARA)is used to determine outpatient surgery eligibility.

Iowa discharge disposition score

A group from the University of Iowa sought to create a predictive calculator for discharge disposition for 107,300patients undergoing TJA procedures using retrospective NSQIP data from 2011–2013 [2]. Researchers compared patients discharged to home versus those discharged to a facility using all variables with a univariate P-value <0.1 in a multivariate logistic regression model to determine risk factors. Values for each variable were assigned a score, and a total score was calculated for each patient. From their analysis, it was found that patients who were discharged to a facility were generally older (70.9 vs 64.3 years, P < .001), female (69.5% vs. 55.7%, P < .001), had elevated ASA class (1 and 2 vs 3 and 4, P < .001) and were more likely functionally dependent before their procedure (3.8% vs 1.1%, P<.001). Predictors of postoperative facility discharge also included increased age, non-elective THA for fracture, and living somewhere besides home before time of surgery. Additionally, the facility discharged group had a complication rate 3 times higher than those home discharged (25.5% vs 8.2%, P<.001), and had a 30-day mortality rate 10 times higher than those home discharged (3.9% vs 0.3%, P<.001). Weighted sores were assigned based on the odds ratio of the five preoperative variables selected, and the authors found that a score of 40 had a 75% chance of discharge to SNF while a patient with a score of 80 had a 99% risk of going to a SNF. Iowa’s discharge disposition tool has not been publicly verified.

Cleveland Clinic calculator

The Cleveland Clinic’s Department of Orthopedic Surgery designed a tool to predict appropriate discharge destinations in order to decrease the use of post-acute care facilities[23]. Their aim was to better prepare families for home support and save on healthcare costs associated with post-discharge care. The authors identified preoperative factors in a retrospective review of approximately 517 TJA cases(between 2005 to 2007) relevant to discharge opportunity within a single institution. These factors were used to build a Predicting Location after Arthroplasty Nomograph (PLAN). Of the 17 variables used in the model, the following seven had statistically significant associations for their independent impact on discharge destination: age, sex, type of arthroplasty procedure, heart disease, diabetes, chronic obstructive pulmonary disease, and the availability of caregiver support. There have yet to be any external studies conducted to validate this tool.

Risk Assessment and Prediction Tool

The Risk Assessment and Prediction Tool (RAPT) was initially developed in 2003 from a cohort of 650 Australian TJA patients investigating the likelihood of prolonged inpatient rehabilitation [24]. The tool has since been modified and applied to American patients in response to bundled payment programs with the intent to optimize discharge disposition for 767 TJA, spinal fusion, and cardiac valve replacement patients (between 2014 to 2015). Patients complete a pre-operative questionnaire with six questions[25]. Scores below six were deemed to be a high risk to not discharge home and those above nine were deemed low risk. The authors of this tool found that in the total joint cohort, scores >9 to be highly correlated discharging a patient home versus scores <3 with 22-times increased odds of discharge to a post-acute care facility (OR = 22.4, 95%CI 10.22, 49.12).

In a 2019 external study to validate the RAPT, retrospective data was collected on 1024 elective TJA patients from 2016–2017 at a single institution [26]. The authors found the tool had an overall predictive accuracy for home discharge of 88%, with predictive accuracies of tool highest at scores <3 and >7 (83–100% and 89–98% respectively. Their analysis found that the patient’s expected discharge location had the strongest correlation with actual discharge disposition (OR = 12.81, CI 7.37, 22.24). The overall findings showed that higher RAPT scores correlated with increased odds of being discharged home (OR = 1.59, CI 1.43, 1.77).

Penn Arthroplasty Risk Score

The University of Pennsylvania’s Departments of Orthopedic Surgery and Anesthesiology–Critical Care developed an algorithm to determine what TJA patients will require post-operative critical care. It was built off of a preexisting institutional model used that was used to stratify preoperative risk specifically for THA patients. The Penn Arthroplasty Risk Score (PARS) was developed to better determine which patients would require critical care following TJA procedures[27]. The PARS began from retrospectively analyzing preoperative and intraoperative variables from 1594 TJA patients at their institution between 2012 and 2013. While 35.2% of the PARS model consists of preoperative variables, they determined certain intraoperative factors were more predictive for a patient requiring critical care. After comparing risk factors between TJA patients requiring critical care and those that did not, the PARS was constructed around the top 5 risk factors with statistical significance in the critical care cohort[28]. A patient’s PARS is calculated from these variables, with the three cardiothoracic risk factors worth one point each, while estimated blood loss over 1000 ml and intraoperative vasopressor use both worth two points. A score of zero gives the baseline probability of requiring a critical care bed of 7%, and scores between 1 and 7 giving an increasing estimate from 13.2% up to 91.1%. When applied more broadly, the same authors retrospectively reviewed their outcomes of 704 consecutive TJA patients using PARS scores to predict discharge status and found fewer patients going to post-acute care facilities after discharge (63% vs. 74%, P =0.002) [29].

Outpatient Arthroplasty Risk Assessment

The Department of Orthopedic Surgery at Indiana University developed the Outpatient Arthroplasty Risk Assessment (OARA) tool to better identify patients that can safely undergo TJA procedures with same day discharge[30]. Their team identified nine risk factor categories that were given a proprietary weighted score with a certain range of points possible to total into a patient’s final OARA score. They determined from their initial clinical application that scores below 59 (out of 1960 possible) were considered safe for early discharge.

980 TJA cases between 2011 and 2016 were analyzed to calculate their retrospective OARA score[30]. These scores were then used to test OARA’s accuracy, given the aforementioned safety threshold score of 59. The mean OARA scores were close for same day discharge (22.2) and next day discharge (30.7) before more sharply increasing for each subsequent day (i.e. 45.3 for discharged patients on day two and 62.6 for day 3 discharges). Patients with an OARA score less than the cutoff of 59 were 2 times more likely to be discharged early compared to scores above the cutoff ([1.4, 2.8], P< 0.001). When comparing their tool against the ASA Physical Status Classification System (ASA-PS) and the Charlson comorbidity index (CCI), they found the PPV of their OARA tool to be 81.6%, significantly higher than both the ASA-PS (56.4%, P< 0.001) and the CCI (70.3%, P = 0.002). Same day discharge patients within this cohort had a 1.8% readmission rate and next day discharges saw a readmission rate of 2.4%.

Two years after OARA’s initial publication, the authors delivered an update to their model after a larger sample was analyzed using OARA[31]. 1785 TJA cases between 2011 and 2018 were retrospectively given OARA scores, and its performance was compared to actual patient outcomes. Performance variables included safe outpatient selection (positive predictive value (PPV), specificity, false positive rates) and comprehensive outpatient selection (negative predictive value (NPV), sensitivity, false negative rates). While 91.5% of same-day discharges had a score below the OARA threshold score of 59 (χ²= 29.333, P< .001), scores below 79 were found to represent 98.8% of all same-day discharge patients (χ²= 28.164, P< .001), giving the range of 0 to 79 a PPV approaching 100%. However, both the 59 and 79 score cutoff ranges showed low NPVs (28.1% and 17%, respectively), low sensitivities (11.4% and 10.7%), and high false negative rates (88.6% and 89.3%). The authors contribute these findings to OARA’s design to “err in the direction of medical safety” to conservatively identify the patients who can safely have a same-day discharge.

A group at NYU applied OARA in a retrospective study of 322 THA patients to determine the validity of the OARA scoring method[32]. They found a PPV of 86.1%for same-day discharge patients, with the PPV remaining relatively constant across all OARA scores in this cohort (which they attribute to their rigorous criteria and protocols that are typical in same-day discharge cases). The study’s standard discharge cohort (targeted for next-day discharge or lengths of stay within 2 days) showed a peak NPV of 88% at an OARA score around 50 with PPV staying in the 30–40% range across all OARA scores. With a high same-day discharge PPV and high standard cohort NPV, the study demonstrated that OARA is better applied as a screening tool to identify subjects at risk for failing to safely undergo early discharge.

Discussion:

A variety of risk calculators have been developed to assess patient risk for surgical events or complications both before and after total joint arthroplasty. Knowledge of a patients risk for readmission and potential discharge disposition is important to surgeons, hospitals, administrators and payors. Prospectively, this data may allow surgeons to intervene and change the health of the patient or modify appropriate risk factors to help prevent readmission and complications. Retrospectively, it is important to risk stratify cohorts based on perioperative characteristics to help guide discharge plans to limit readmissions, especially in riskier populations. The ability to predict postoperative disposition is increasingly important as TJA transitions to an outpatient procedure.

Our review identified 10 different risk assessment tools used to determine hospital readmission risks and discharge destination after TJA. While unique in many regards to input variables, all of the prediction instruments incorporate patient age, while most include gender, BMI and smoking status (Table 1). We found that differences between stratification tools become starker when evaluating the subtypes and relative value of certain comorbidities that contribute to each equation (Table 2). Modifiable risk factors known to affect readmission are of importance when calculating patient risks surrounding the TJA episode [33]. Knowing what modifiable factors are prevalent in a surgeon’s patient population and addressing them preoperatively are essential to mitigate the risks of complications [34]. While the RRAT tool exclusively focuses on modifiable risk factors, all but two calculators apply modifiable risk factors to some degree (BMI, smoking, comorbidities) in their algorithm (Table 2). However, both the Iowa discharge tool and RAPT are primarily weighted from a patient’s preoperative functional status and living arrangements, which are not easily altered.

Table 1.

Comparison of input variables used in selected risk stratification tools used to predict readmission and discharge status after total joint arthroplasty.

Risk Calculator Inputs	ACS (1)	AJRR (2)	Duke (3)	Ortho Cincy (4)	RRAT (5)	Iowa (6)	Cleveland (7)	RAPT (8)	PARS (9)	OARA (10)
Age	x	x	x	x		x	x	x
Sex	x	x		x		x	x	x
Race		x
BMI	x	x3		x	Obesity		x
Smoking Status	x		x	x	x
Insurance Status		x	x	x
Elective or Emergent	x					x
Procedure	CPT code		x4				Revision?		X2
ASA	x		x			x
Functional Dependence	x4				Assistance	x2	x5	x4
Recent Complications	x3			TJA w/in 12 mo
Comorbidities*	x9	x30	x8	x7	x4		x5		x3	x9
Psychosocial	x2	x2	Depression	Any diagnosis	Neurocognitive.
Discharge Factors				x

^*see Table 2 for associated comorbidities

1. ACS Dependence: Functional Status, mobility aide usage, home support available, fall history

ACS Recent Complications: CHF (w/in 30 days), ascites (w/in 30 days), sepsis (w/in 48 hours)

ACS Psych: Cognitive Status, competency at admission

2. AJRR BMI: Height, weight, obesity

AJRR Psych: Depression, Psychoses

3. Duke Procedure: Which joint, surgery duration, post-op Hemoglobin, post-op BUN

6. Iowa: Dependence: Functional Status, Location before surgery

7. Cleveland: Dependence: # entry steps, bedroom location, home assistance, home location, weight-bearing status

8. RAPT Dependence: Walking distance, mobility aide usage, home assistance available, community support available

9. PARS Procedure: Estimated blood loss, intraoperative vasopressors

10. OARA: While “General Medical” is listed as a category, no details have been published as to what comprises this category

While one risk factor alone may not contraindicate surgery, the additive effect of multiple risk factors together increases the cumulative risk of readmission or morbidity. The current calculators available apply different methodologies to weigh the relative contributions of comorbid and demographic factors to produce a quantitative probability of risk. Population comparisons can further prove helpful in assessing the relative risk of a patient to their counterparts. This feature is only used in the ACS-NSQIP tool, where a given patient’s risks are calculated relative to the average patient. The ACS surgical risk calculator has demonstrated good to fair accuracy in predicting discharge to post-acute care facility, but has poorer utility in predicting 90-day complications, unplanned return to surgery, and post-surgical complications such as DVT and periprosthetic joint infection [35]. The AJRR TJR risk calculator attempted to assess the relative risk of complications compared to those with similar or equal demographics without comorbidities, potentially showing how both modifiable and non-modifiable comorbidities multiplicatively increase one’s risk for 90-day complication. However due to poor reproducibility this tool is not currently in use. A new tool from the AJRR is currently being developed.

Discharge disposition assessments can be used to more appropriately select a patient’s correct discharge location (home vs. facility). In the age of rapid recovery TJA, this may help surgeons designate patients who are not candidates for early or same-day discharge. This is particularly important for ambulatory surgical centers, where patients cannot be directly admitted to potentially quality for post-acute care facility placement[36]. Out of the five discharge tools included in our review, the RAPT and OARA calculators were independently verified. RAPT was unique in that it was derived from subjective assessments by the treating physician and their belief whether the patient should be discharged to a facility or not. The tool defined low, medium, and high-risk ranges based off of the differential answers between patients discharged home and those discharged to a facility. One unique factor to this tool found on external validation was that the patient’s actual preference on discharge disposition was found to have strongest correlation to actual discharge destination in addition to RAPT score[26]. However, in the age of cost control and bundled care, this decision-making is not entirely up to the patient.

The tools in this study were all developed prior to the current move to outpatient surgery. The only tool that examines decision making for outpatient surgery is the OARA. While OARA aimed to score a patient’s risk to determine who can safely undergo outpatient arthroplasty, the weighted variables included in their nine general categories are proprietary and not readily published [31]. Although the tool’s PPV has been internally and externally validated (both in retrospective studies), the outpatient cutoff scores were retrofitted to retrospective patient data and should be validated clinically in prospective studies.

The other three discharge tools (Iowa, Cleveland Clinic, and PARS) built their models using select variables that showed strong associations to discharge disposition. The most noticeable difference between the tools is that the Cleveland Clinic tool used more expansive input options, compared to two functionality variables for the Iowa tool [2,23]. The PARS tool uses intraoperative variables and none of these calculators have been validated externally. While the simplicity of the Iowa group’s calculator is appealing due to the limited number of input variables and the Cleveland Clinic tool is more comprehensive, both prove impractical to use preemptively due to the variable weighting of certain factors and the non-modifiable nature of the demographic and comorbid factors used. PARS remain an outlier among these in using intraoperative variables to determine if patients will need ICU critical care, which invariable affects subsequent discharge decisions.

Choosing between risk calculators depends on the demographic and comorbid factors available to the surgical team before proceeding with TJA and involves compromises (Table 3). Although the exact equations for many of the calculators are not readily obtainable, the methods of how they were built are often published with the calculator. While certain calculators consider a large number of potential factors in calculating a risk relative to a baseline population (ACS, AJRR), others use regression analysis to retroactively assign risks based on associations between certain patient demographic and disease state characteristics and readmission or discharge occurrences (Duke, OrthoCincy, RRAT, Iowa, Cleveland Clinic, PARS, OARA). If a surgeon seeks to determine a patient’s risk for future operations, calculators using intra-operative and post-operative variables (Duke, PARS) cannot aid in determining the most accurate risk probabilities prior to the TJA event, potentially skewing a patient’s calculated risk from a pre-operative risk-mitigation standpoint. Two calculators in particular, ACS and RAPT, base their risk calculations from data on surgeries outside of TJA and orthopedics, making them less desirable in attempts to discover risks based solely on arthroplasty procedures.

Limitations

While risk calculators show promise in predicting arthroplasty readmission, outcome and disposition risks, many limitations exist. The first shortcoming with many of the risk stratification tools is that they lack external validation to prospectively applied patient data at other institutions. Only four of the ten tools have any type of validation published in the literature. This limitation is compounded by the fact that most of the formulas used to generate each calculator are not freely published, so deconstructing the process for each risk generator can be difficult when comparing one calculator to another. Furthermore, some of these prediction tools utilize retrospective NSQIP data from the earlier half of the 2010s (ACS, Iowa), when discharge rates to post-acute care facilities were more common, as rapid recovery protocols were not yet fully in place. Thus, the inherent data used to generate particular calculators may not reflect the current healthcare landscape in which enhanced recovery pathways drive post-surgical care. A third limitation is that no study applies a single set of patient data to multiple calculators to independently determine if one prediction tool has better accuracy or precision pertaining to readmissions and/or discharge disposition. In reality, this would very difficult to accomplish for the reasons listed above with regards to the lack of transparency for each calculator’s individual formula and differences between contributing variables. A fourth limitation in the utilization of these calculators is because of the rapidly evolving pace of arthroplasty. Outpatient TJA and same-day discharges are becoming commonplace and were not widely utilized when these calculators were generated. It is unknown how many of these calculators are currently utilized in general orthopaedic practice. Lastly, it is possible that our method of investigating only readmission and discharge risk calculators may have missed other types of tools or strategies available to surgeons and policy makers to better stratify potential arthroplasty patients.

Conclusions

No risk stratification tool for total joint arthroplasty appears perfect and can be applied in all situations. Nonetheless, each of the tools reviewed can have utility based on their intended application. Even though AJRR and ACS may beneficially compare patient’s risk to standard populations with the most comprehensive inputs, their validation studies show poor predictability [24, 26].The other three readmissions-focused calculators have not been validated, however, OrthoCincy uses a higher quality of inputs to better encompass a patient’s risk. The RAPT and Cleveland Clinic disposition tools allow for more granularity when accounting for a patient’s functional status, with their differences coming from comorbidity inputs (simplicity versus breadth). The OARA may be particularly help in the screening patients on appropriateness for outpatient joint replacement.

The extensive variation among TJA risk calculators underscores the need for tools with more individualized stratification capabilities and verification. The transition to outpatient and same day discharge TJA may preclude or change the need for many of these calculators. Further studies are needed to develop more streamlined risk calculator tools that predict readmission and surgical complications..