Predicting Delayed In-Hospital Recovery of Physical Function After Total Knee Arthroplasty

Highlights

• •Preoperative fitness is associated with postoperative recovery of physical function.
• •Predictors include handgrip strength, walking aid indoor, and the de Morton Mobility Index.
• •Distinguishing between high- and low-risk patients remains difficult.

Abstract

Objective

To identify patients at high risk of delayed in-hospital functional recovery after knee replacement surgery by developing and validating a prediction model, including a combination of preoperative physical fitness parameters and patient characteristics.

Design

Retrospective cohort study using binary logistic regression.

Setting

University hospital, orthopedic department.

Participants

260 adults (N=260) (≥18y) with knee osteoarthritis awaiting primary unilateral total knee arthroplasty and assessed during usual care between 2016 and 2020.

Intervention

Not applicable.

Main Outcome Measures

Time to reach in-hospital functional independence (in days), measured by the modified Iowa Level of Assistance Scale. A score of 0 means completely independent. Potential predictor variables are a combination of preoperative physical fitness parameters and patient characteristics.

Results

Binary logistic regression modeling was applied to develop the initial model. A low de Morton Mobility Index (DEMMI), walking aid use indoors, and a low handgrip strength (HGS) were the most important predictors of delayed in-hospital recovery. This model was internally validated and had an optimism-corrected R² of 0.07 and an area under curve of 61.2%. The probability of a high risk of delayed in-hospital recovery is expressed by the following equation:

$P_{highrisk}=(1/(1+e^{(-(2.638-0.193\times DEMMI+0.879\times indoorwalkingaid-0.007\times HGS\left)\right)}))\times 100\%$.

Conclusions

The model has a low predictive value and a poor discriminative ability. However, there is a positive association between preoperative physical fitness and postoperative recovery of physical function. The validity of our model to distinguish between high and low risk, based on preoperative fitness values and patient characteristics, is limited.

Currently, treatment for end-stage knee osteoarthritis (KOA) follows a stepped care process that starts with a conservative approach as the first step and ends with total knee arthroplasty (TKA) as the last step.¹ TKA is considered the criterion standard for treatment of end-stage KOA.² End-stage KOA often results in limitations of activities of daily living (ADL), work, and leisure.^3,4 In 2021, 21,444 TKA surgeries were performed in The Netherlands.⁵ The incidence of TKA increases with age and is higher in women.⁵ Because of the aging population, the number of TKA surgeries is expected to increase in the coming years.

To cope with the increasing burden of hospital admissions, recovery pathways such as enhanced recovery after surgery, joint care, fast-track and rapid recovery approaches have been developed.⁶ The aim of these care pathways is to reduce postoperative physical and psychological stress for all sorts of patients, thereby decreasing time to functional recovery and reducing the length of hospital stay (LOS). Delayed in-hospital functional recovery results in a longer hospital stay and may lead to increased risks of postoperative complications.⁷

At the Maastricht University Medical Center+ (MUMC+) in Maastricht, The Netherlands, new recovery pathways have reduced the time to reach independent functional recovery, from a median (range) of 4 (2-5) to 2 (1-8) days in patients after elective knee and hip replacement surgeries. LOS was reduced from 7 (5-11) to 4 (3-12) days.⁸

Patients are considered functionally recovered if they are able to independently perform fundamental ADLs, such as transferring and walking.9, 10, 11 Performing these functional activities independently can be effectively evaluated in orthopedic patients by the validated modified Iowa Level of Assistance Scale (mILAS).¹⁰ Although multiple modifications have already been implemented during the in-hospital phase, further optimization of the preoperative phase is required. In elective TKA procedures, surgery is usually planned within 4-6 weeks after diagnosis. In this timeframe, interventions can be implemented to better prepare these patients for surgery, mainly by enhancing physical fitness.¹² In nonorthopedic fields, rehabilitation programs maintain or even improve physical function and reduce the risk of worsened function while patients await surgery.13, 14, 15, 16 Enhanced functional capacity before surgery also serves to reduce postoperative pain, prevent complications, and reduce hospital LOS in patients after TKA and hip arthroplasty.^13,17

Several systematic reviews have shown positive relationships between pre- and postoperative physical function in TKA patients.^11,18 The literature shows that patient characteristics and variables of physical fitness^9,11,19 have been found to be prognostic factors for in-hospital functional recovery after TKA. Preoperative physical function is an important predictor of physical function after surgery in TKA patients.^11,18 Prehabilitation has also been shown to improve physical function before and within the first week after TKA.^13,20, 21, 22, 23, 24, 25, 26, 27 However, increasing numbers of TKA procedures make it impossible and financially ineffective to prehabilitate every patient. It has, therefore, been suggested to focus on patients at high risk of delayed recovery after surgery, as these patients and the health care system would probably benefit the most. Although risk stratification would enable identification of high-risk patients, there is currently a lack of consensus on the definition of high risk²⁸ or the ability to determine whether a patient is at high risk of delayed recovery before surgery. Selection of participants,^13,29 choosing the right combination of predictors, and outcome measurements are still challenging.¹⁸

The aim of this study was, therefore, to identify patients’ risk of potential delayed in-hospital recovery of physical function after TKA. The primary goal was to develop a prediction model based on a combination of variables of physical fitness and personal characteristics, which led to the following research question:

Is it possible to identify TKA patients at high risk of delayed in-hospital functional recovery by developing a prediction model based on a combination of variables of physical fitness and personal characteristics?

Methods

Design

In this retrospective cohort study, we developed a model to predict whether patients will achieve in-hospital functional independence. Patients entered the cohort when TKA surgery was planned between June 2016 and May 2020 at MUMC+. Procedures of data collection complied with the Declaration of Helsinki and were approved by the Ethical Review Board of MUMC+ (registration number 2020-1337). Because of the retrospective study design, in which data were collected during usual care, informed consent was not applicable.

Participants

Patient data were collected during usual care, which involved patients being routinely assessed before receiving a primary unilateral TKA for KOA. Functional independence was required for all patients to return home after surgery. Patients who were scheduled to go to a rehabilitation clinic for further recovery after surgery were excluded, as they may leave the hospital before reaching functional independence.

Procedure

Four to 6 weeks before surgery, patients were invited to visit the physical therapy department of MUMC+ for preoperative assessment of their physical function. During this assessment, patient characteristics were collected and patients performed several tests to assess their level of preoperative physical function (table 1). Additionally, patients received information about in-hospital physical therapy and were advised to stay active until surgery.

Table 1.

Preoperative assessments

Test	Domain
TUG30	Balance, walking ability
2MWT31	Self-paced walking ability and functional capacity
DEMMI32	Physical function
HGS33	Overall muscle strength

Abbreviations: TUG, Timed up and go; 2MWT, 2-minute walking test; DEMMI, de Morton Mobility Index; HGS, handgrip strength.

When surgeries were planned in the morning, physical therapy started in the afternoon of the same day. The physical therapy treatment goal was regaining independent ADL functioning. All sessions contained protocolled mobilizing exercises and practicing transfers and gait training with a walking aid that was adapted to the patient's ability. Patients received physical therapy twice daily until reaching functional independence.

Primary outcome

The primary outcome was the time to reach functional independence (in days) as measured by the mILAS, which has a high inter-rater reliability (intraclass correlation coefficient [ICC] 0.96; 95% confidence interval [CI] 0.93-0.98)³⁴ and is a valid and responsive measure to determine the level of independence in patients after TKA.¹⁰ The mILAS assesses the amount of assistance needed to safely perform the following ADLs: transition from supine to sit, sit to supine, and sit to stand; walking 4 meters; and climbing stairs.¹⁰ The score per item ranges from 0 (independent) to 6 (not able due to medical or safety reasons). A total score of 0 points indicates functional independence. The continuous probability resulting from the prediction model was dichotomized in an approximately 30%:70% ratio, based on previous literature analysis.³⁵ Patients who reached functional independence within 2 days were categorized as low risk (70%), while those who needed more than 2 days to reach functional independence were categorized as high risk (30%).

Potential predictor variables

The following demographic variables were assessed during the preoperative screening: age in years; sex (male/female); body mass index (BMI) in kg/m²; American Society of Anesthesiologists score (range 1-3), where a higher score indicates lower fitness for surgery³⁶; Charnley-score (A-C) in which patients score A-B or C, meaning they already have arthrosis in 1 (A) or both knees (B1), already have a total knee replacement (B2), or have multiple joints affected by osteoarthritis or a chronic disease that impairs their quality of life (C)³⁷; indoor and outdoor use of a walking aid (cane, crutch, walking frame, walker, wheelchair, or scooter); and prevalence of comorbidities (cardiac, pulmonary, vascular, neurologic diseases, or diabetes mellitus).

Timed Up and Go test (TUG) was used to assess functional mobility.³⁰ It measures the time (in seconds) it takes for the patient to stand up from a chair, walk 3 meters as quickly as possible, turn, walk back, and sit down in the chair (walking aid permitted). The TUG is a common performance-based outcome measure of physical function in patients with KOA and TKA.³⁸ It has excellent test-retest reliability (ICC=0.97).³⁹

Handgrip strength (HGS) was used to measure maximal HGS (in kilograms) to indicate overall muscle strength.³³ HGS was measured using a handheld dynamometer (JAMAR Hand Dynamometer, Patterson Medical). The patient had to squeeze the dynamometer as hard as possible while sitting on a chair without armrests, feet resting flat on the floor, elbows flexed at 90°, forearm in a neutral position, and wrist in 0°-30° extension.⁴⁰ It is a common performance-based measurement and an important predictor of functional limitations in older adults.^33,41 The test has an excellent test-retest reliability (ICC=0.98).⁴²

The 2-minute walking test was used to assess normal walking speed and endurance capacity by measuring the walking distance in meters.³¹ During the test, patients had to walk as far as possible within 2 minutes (walking aid permitted). The test-retest reliability is excellent (ICC=0.97).⁴³

de Morton Mobility Index (DEMMI) was used to assess the level of independence in basic motor activities of older patients.³² The measurement consists of 15 items. Each item is scored on a 2- or 3-point scale (0-1 and 0-1-2), which results in a score of 0-19 points. A higher score indicates better independent mobility.⁴⁴ The DEMMI is a reproducible, valid, and feasible instrument in older patients with KOA⁴⁴, with an inter-rater reliability of ICC=0.85; 95% CI 0.71-0.93.⁴⁴

Sample size

The sample size was based on having at least 5 events per candidate predictor for the logistic regression analysis.⁴⁵ The hypothesis was that approximately 30% of the patients would be at risk of delayed recovery. The literature showed that 30%-60% of patients aged 70 years and older face reduced independence after hospitalization, regardless of whether they were fully independent before admission.^35,46 Including all 14 predictor variables in the model required a sample size of at least 234 (5×14×(10/3)) patients.⁴⁵ Taking into account a dropout rate of at least 10%, 257 participants were needed.

Statistical analysis

Data analysis was performed using SPSS version 25.0^a (IBM, New York, USA) and R Statistical Software version 4.04^b (R Core Team 2021). Continuous variables are presented as means with standard deviation for normally distributed data and as medians with first and third quartiles for non-normally distributed data. For categorical variables, we used frequencies with percentages. Visual inspection of histograms and the Kolmogorov-Smirnov test were used to check for normality (P≤.05). Data were checked for completeness. We used multiple imputation to impute missing values, with predictive mean matching as the model for continuous variables.⁴⁷ We set the number of imputations to 5.⁴⁸ A sensitivity analysis was performed to check that imputation would not lead to very different results. In the statistical analysis, the number of days required to reach mILAS 0 was used as a dichotomous variable to be able to distinguish between patients with and without delayed recovery of function.

Model development

We used logistic regression with the dichotomized mILAS as the outcome variable. Patient characteristics and performance tests were assessed as candidate predictors. The presence of comorbidities and the use of a walking aid were entered into the model as dichotomous variables (yes/no); for the American Society of Anesthesiologists score and Charnley-score, 2 dummy variables were used. Backward stepwise elimination based on the Wald test was used for variable selection. An alpha of 0.20 was used, as recommended by prediction modeling guidelines.⁴⁹ Predictors that remained in at least 3 of the 5 imputations were selected and entered into the final model. Multicollinearity between candidate predictors was assessed using the variance inflation factor, and linear associations with log odds of the outcome for continuous variables were checked using the Box-Tidwell test.

Internal validation

The model was internally validated using bootstrapping (B=1000). A shrinkage factor was estimated to adjust the model coefficients. After shrinkage, the model intercept was re-estimated. In addition, we computed an optimism-adjusted measure of model performance. The overall performance of the model was quantified using Nagelkerke's R² and the Brier score. The ability of the model to discriminate between patients at a low and high risk of delayed recovery of physical function was expressed as the area under the receiver operating characteristic curve. The bootstrap method, which is used for internal validation, gives insights into estimates of the optimism in predictive performance of the model and gives a shrinkage factor to adjust for overfitting. Adjusted performance of the model gives a better view of what can be expected if the model is used in other patients from the same population.⁵⁰ The agreement between predicted probabilities and observed frequencies of the outcome was assessed by evaluating the calibration plot. A Hosmer-Lemeshow goodness-of-fit statistic was computed to evaluate the calibration. Additionally, sensitivity, specificity, positive predictive value, and negative predictive value were calculated. Cutoffs were based on the Youden's J statistic and the highest sensitivity.

Results

In total, 260 patients scheduled for primary unilateral TKA underwent preoperative screening and returned home after surgery. These patients had a median age of 67 (Q1–Q3: 61-74) years and a median BMI of 30.7 (Q1–Q3: 26.6-34.4) kg/m². About half were female (54.2%). The median time to reach functional independence was 2 (Q1–Q3: 1-3) days. Data concerning the main outcome of reaching functional independence was lacking for 15 patients. A total of 175 patients (67%) reached functional independence within 2 days and were categorized as low risk. In 70 patients, reaching functional independence took longer, and they were categorized as high risk. For 15 patients, functional independence at discharge remained unclear (table 2). Table 3 shows the preoperative physical performance test scores for the high- and low-risk groups; there were significant differences between the groups for the TUG and DEMMI.

Table 2.

Patient characteristics

Age (y)	67 (61-74)
Sex (n)
Male	119 (45.8)
Female	141 (54.2)
BMI (kg/m2)	30.7 (26.6-34.4)
ASA score (n)
1	32 (12.3)
2	171 (65.8)
3	57 (21.9)
Charnley-score* (n)
A	127 (48.8)
B	115 (44.2)
C	15 (5.8)
Days to mILAS 0* (d)	2 (1-3)
Days to mILAS 0* (n)
≤2	175
>2	70
Comorbidities (n)
Yes	129 (49.6)
No	131 (50.4)
Walking aid (n)	Indoors	Outdoors
Yes	19 (7.3)	64 (24.6)
No	241 (92.7)	196 (75.4)

NOTE. Data are presented as median (first quartile to third quartile) or frequencies (percentage).

Abbreviations: y, year; n, number; kg, kilogram; m, meter; d, days; BMI, body mass index; ASA, American Society of Anesthesiologists; mILAS, modified Iowa Level of Assistance Scale.

^⁎Variable contains missing data.

Table 3.

Preoperative physical performance tests

Test	Low Risk* (n=175) ≤2 days to reach mILAS=0	High Risk* (n=70) >2 days to reach mILAS=0	P Value†
TUG (s)	8.02 (6.72-9.97)	9.05 (6.92-12.13)	.035
HGS (kg)	32 (26-42)	30 (22-38)	.091
2MWT*‡ (m)	138.0 (118.0-163.0)	134.5 (106.5-156.8)	.232
DEMMI* (score)	18 (17-19)	18 (16-18)	.003

NOTE. Data are presented as median (first quartile to third quartile).

Abbreviations: mILAS, modified Iowa Level of Assistance Scale; TUG, timed up and go; HGS, handgrip strength; 2MWT, 2-minute walking test; DEMMI, de Morton Mobility Index.

^⁎Variable contains missing data.

^†P value <.05. The Mann-Whitney U test was used to compare means.

^‡Data are normally distributed.

Model development and internal validation

A binary logistic regression was performed with predictors that were present (DEMMI, indoor walking aid, and HGS) in at least 3 of the 5 imputation equations. The regression resulted in a pooled model, which is shown as the initial model in table 4. The optimism-corrected R² was 0.07. A shrinkage factor of 0.90 was estimated. After shrinkage, the model intercepts were re-estimated and are shown in the column of the internally validated model (table 4). After internal validation, the probability of having a high risk of delayed in-hospital recovery was represented by the following equation:

$$P_{highrisk}=(1/(1+e^{(-(2.638-0.193\times DEMMI+0.879\times indoorwalkingaid-0.007\times HGS\left)\right)}))\times 100\%$$

Table 4.

Initial and internally validated prediction models

	Initial Model			Internally Validated Model
Variables	Regression coefficient	Odds ratio (95% CI)	P value	Regression coefficient
Constant	3.029	-	.074	2.638
DEMMI	–.214	.808 (.655-.996)	.046	–.193
Walking aid indoors	.978	2.660 (.802-8.823)	.109	.879
HGS	–.008	.992 (.964-1.020)	.567	–.007

Abbreviations: CI, confidence interval; DEMMI, de Morton Mobility Index; HGS, handgrip strength.

The AUC was 63.7% (95% CI=0.559-0.714) (fig 1). Variance inflation factor statistics were 1.251 for indoor walking aid, 1.147 for HGS, and 1.409 for DEMMI. There were no linear associations with the log odds of the outcome for HGS (P=.638) and DEMMI (P=.555). The optimism-corrected AUC was 61.2%.

Fig 1.

Receiver operation curve showing the ability of the model to discriminate between patients at a low and high risk of delayed in-hospital recovery of physical function after TKA. Expressed in an AUC of 63.7% (95% CI=0.559-0.714).

Table 5A shows a selection of probability cutoff values based on the best Youden's Index, with a probability threshold of 0.34, and based on the highest sensitivity, with a probability threshold of 0.19 (table 5B). The sensitivity analysis showed no large differences between the original and imputed datasets. The Hosmer-Lemeshow goodness-of-fit test was nonsignificant (P=.687).

Table 5A.

Cutoff values based on best Youden's index

		Actual
		Delayed recovery	Normal recovery
Predicted	Delayed recovery	28 (TP)	27 (FP)	PPV 51%
	Normal recovery	44 (FN)	161 (TN)	NPV 79%
		Sensitivity 38.9%	Specificity 85.6%

Abbreviations: FP, false positive; FN, false negative; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive.

Table 5B.

Cutoff values based on highest sensitivity

		Actual
		Delayed recovery	Normal recovery
Predicted	Delayed recovery	65 (TP)	145 (FP)	PPV 31%
	Normal recovery	7 (FN)	43 (TN)	NPV 86%
		Sensitivity 90.3%	Specificity 22.9%

Abbreviations: FP, false positive; FN, false negative; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive.

Discussion

In this study, we developed and internally validated a prediction model to identify patients at risk of delayed in-hospital recovery after TKA surgery based on preoperative parameters. The patient characteristics of the included patients were in line with the findings of other studies of TKA patients.^9,13,19 The primary outcome of the study was the time needed to reach in-hospital functional independence (in days), measured by the mILAS. Of the included population, 175 patients (67%) reached functional independence within 2 days and were defined as low risk. Seventy patients took longer and were defined as high risk.

Delayed (>2 days) recovery of in-hospital functional independence was predicted by a combination of the DEMMI score, the HGS, and the preadmission use of a walking aid indoors. All other predictor variables turned out to be nonsignificant. The model was able to distinguish between being at low or high risk of delayed in-hospital recovery of physical function. However, the optimism-corrected AUC was 61.2%, which was considered poor discrimination.⁵¹ Using a walking aid indoors before surgery turned out to be the most important predictor of delayed in-hospital recovery of physical function, increasing the predicted probability by at least 20%. If patients did not use a walking aid indoors, the DEMMI score was the most important predictor. Patients with a score of 16 points or lower on the DEMMI are always at risk.

The literature shows that patient characteristics and variables of physical fitness^9,11,19 are prognostic factors for in-hospital functional recovery after TKA. The clinical purpose of using a prediction model is to select patients at risk of delayed recovery to better prepare them physically for surgery (prehabilitation). These models are also intended to hopefully reduce the time of functional recovery for similar patients in the future. Hence, we need working models with appropriate cutoff scores. Our model was only able to predict a small proportion of the outcome (Nagelkerke R² 0.089). The difficulty of developing a prediction model that works well has already been discussed in the literature.^18,19,52

Our study yielded 2 possible cutoff scores. The first option was based on the best Youden's index and had a specificity of 85.6%. Using this model, the number of false-negative patients was quite high: These scores resulted in 44 patients with an actual prolonged recovery time being assigned to the low-risk category and, thus, receiving no preoperative intervention. Furthermore, 27 patients who did not actually need a prolonged time to recover were assigned to the high-risk category; based on the model, these patients would have received prehabilitation. In the second option, based on a higher sensitivity of 90.3%, the number of false-negative patients was only 7. However, in this option, 145 patients were incorrectly classified by the model as positive and, thus, would receive prehabilitation that probably would not reduce their time to functional independence.

An additional aspect that made prediction in this cohort difficult is the fact that before hospital admission, some patients were already scheduled for inpatient rehabilitation based on decreased physical fitness and/or social factors, such as living alone or having little social support in the immediate environment. These patients were, therefore, excluded from the analysis. As a result, the included patients turned out to be very homogeneous (table 3). The remaining group of high-risk and low-risk patients had very similar patient characteristics and only differed significantly on the TUG and the DEMMI.

Strengths

A strength of this study was the large study population. Prediction modeling requires at least 5 events per predictor variable⁴⁵, a condition which was fulfilled in this study. Second, data of the mILAS was gathered during regular health care visits, which best reflects daily practice. Third, both preoperative and postoperative measurements were performed by well-trained physiotherapists with experience with orthopedic patients, ensuring good inter-rater reliability.^34,39,42, 43, 44

Limitations

This study also had several limitations. First, not all patients were included in the analysis because patients who were going to a rehabilitation center or a nursing home were excluded from the analysis. Patients who are not able to live at home independently after surgery are more likely to have delayed recovery of physical function. Second, in our data analysis, missing data were imputed by multiple imputations, which could lead to over- or underestimation of the results. However, missing data cannot be avoided in observational research and can be a cause of missing information or bias.⁴⁸

Despite the large sample size, we were still unable to create a good working prediction model. One possible reason for this is that there are many more potential predictors involved in the perioperative process than we have included. However, we deliberately chose predictors we can influence as physiotherapists. Another reason may be that the patients in both the low- and high-risk groups are very similar. When the differences are so small, developing a model with discriminatory ability is very difficult and often leads to misclassification.

Conclusions

An internally validated prediction model was developed to predict in-hospital delayed recovery of physical function after TKA. The predictive value of the model was low, and its discriminative ability was poor. Thus, whereas it can be concluded that there is an association between preoperative physical fitness and postoperative recovery of physical function, the distinction between high and low risk must be made with caution.

Suppliers

• a.IBM SPSS Statistics for Windows v.28; IBM Corp.
• b.R Core Team (2021), v.4.04; R Foundation for Statistical Computing.