 KNEE

A prediction model for length of stay after total and unicompartmental knee replacement P-H. Ong, Y-H. Pua From Department of Physiotherapy, Singapore General Hospital, Singapore

Early and accurate prediction of hospital length-of-stay (LOS) in patients undergoing knee replacement is important for economic and operational reasons. Few studies have systematically developed a multivariable model to predict LOS. We performed a retrospective cohort study of 1609 patients aged ≥ 50 years who underwent elective, primary total or unicompartmental knee replacements. Pre-operative candidate predictors included patient demographics, knee function, self-reported measures, surgical factors and discharge plans. In order to develop the model, multivariable regression with bootstrap internal validation was used. The median LOS for the sample was four days (interquartile range 4 to 5). Statistically significant predictors of longer stay included older age, greater number of comorbidities, less knee flexion range of movement, frequent feelings of being down and depressed, greater walking aid support required, total (versus unicompartmental) knee replacement, bilateral surgery, low-volume surgeon, absence of carer at home, and expectation to receive step-down care. For ease of use, these ten variables were used to construct a nomogram-based prediction model which showed adequate predictive accuracy (optimism-corrected R2 = 0.32) and calibration. If externally validated, a prediction model using easily and routinely obtained pre-operative measures may be used to predict absolute LOS in patients following knee replacement and help to better manage these patients. Cite this article: Bone Joint J 2013;95-B:1490–6.

 P-H. Ong, BAppSc (PT), Principal Physiotherapist  Y-H. Pua, PhD, Principal Physiotherapist Singapore General Hospital, Department of Physiotherapy, Outram Road, Singapore 169608, Singapore. Correspondence should be sent to Dr Y-H. Pua, e-mail: [email protected] ©2013 The British Editorial Society of Bone & Joint Surgery doi:10.1302/0301-620X.95B11. 31193 $2.00 Bone Joint J 2013;95-B:1490–6. Received 1 November 2012; Accepted after revision 3 June 2013

1490

Hospital beds are a limited resource in constant demand. In the context of knee replacement surgery (total or unicompartmental), it is expected that the requirement for this procedure will increase seven-fold from 2005 to 2030.1 It is important to manage the length of hospital stay for both economic and operational reasons. Early and accurate prediction of the length-of-stay (LOS) allows better bed management and allocation of resources, shapes patients’ expectations, and facilitates discharge planning. Studies in the literature investigating factors associated with LOS after knee replacement have often included mixed samples of hip and knee replacement2-7 and have not performed multivariable analyses or were limited by small sample size for such analyses.4-6,8,9 These studies have tended to focus on calculating the odds of a prolonged LOS (based on an arbitrary cutoff) without systematically developing a multivariable model to predict absolute LOS. To our knowledge, only three studies have attempted to do so.10-12 One such study was published in 1996,10 but subsequent improvements in surgical techniques and post-operative management have substantially shortened the LOS for

current patients, which, in turn, limits the usefulness of the proposed model. Although the other two studies were more recent,11,12 both were concerned with predicting the LOS for inpatient rehabilitation and not the length of the acute hospital phase of care. Thus, the aim of our study was to develop an accurate but simple predictive model for LOS in knee replacement in the acute care setting. To accomplish this aim, we focused on pre-operative variables that are easily or routinely measured in clinical practice.

Patients and Methods Study population. We retrospectively studied a cohort of 1798 patients aged ≥ 50 years who underwent a primary, elective, total or unicompartmental knee replacement for knee osteoarthritis at Singapore General Hospital, between August 2009 and January 2011. We excluded patients who were non-permanent residents of Singapore (n = 42) and we also excluded patients who developed post-operative medical and surgical complications that adversely affected outcomes (n = 37) because these patients were a small heterogeneous group and their protracted LOS was more closely related THE BONE & JOINT JOURNAL

A PREDICTION MODEL FOR LENGTH OF STAY AFTER TOTAL AND UNICOMPARTMENTAL KNEE REPLACEMENT

to the severity of the complications than to the conventional predictors of LOS. For patients with consecutive admissions for knee replacements (n = 110), only data from the first admission were analysed in order to meet the statistical assumption of independence. Accordingly the sample for analysis comprised the remaining 1609 patients who underwent a pre-operative evaluation within five weeks prior to surgery. All patients who underwent knee replacement were managed using a co-ordinated clinical pathway to ensure standardised medical, pharmacological pain management, and rehabilitation care. Spinal/epidural infusions or nerve blocks were not used in this cohort of patients for post-operative pain management. The majority of patients (n = 1128, 70.1%) received patient-controlled analgesia (with a morphine bolus of 1 mg, lock-out time of five minutes, and a maximum dose of 8 mg/hour) for up to 48 hours, while a smaller proportion of patients received either intravenous morphine infusion at a rate of 0.5 to 2 mg/hour or an intraarticular injection comprising 0.5 ml/kg of 1:200 000 adrenaline and 0.5% bupivacaine diluted with 30 ml of normal saline, and 40 g of corticosteroid (triamcinolone acetonide). Oral analgesia was commenced 24 hours after the operation. This included 1 g of paracetamol every six hours and 550 mg of naproxen every 12 hours, taken with 20 mg of omeprazole daily. If a patient was allergic to naproxen, 50 mg of tramadol every eight hours was given instead. All data were collected by physiotherapists and technicians and entered into an electronic registry database in accordance with routine practice at our institution. The Singhealth Institutional Review Board approved the study. Variables. The independent (predictor) variables that we identified are listed in Table I. These variables were chosen based on their theoretical or documented association with LOS.2,3,9,12 Comorbidities were defined according to the 13 comorbid conditions stipulated by Sangha et al.13 In our study, as a result of low prevalence of some conditions (e.g., stomach diseases, anaemia and lung diseases) and no clinically or statistically significant differences in LOS between the various comorbidities, the number of comorbidities was recoded into three categories: i) none; ii) one to three comorbidities; and iii) ≥ four comorbidities. Each patient was interviewed in either English (460 patients, 28.6%) or Mandarin (1149 patients, 71.4%) using Short-Form 36 (SF-36),14,15 of which we used the bodily pain subscale. In order to assess self-reported depression, a single question (Q28) from the SF-36 (‘How much of the time during the past four weeks have you felt downhearted and depressed?’) was used.16,17 Based on a pre-specified classification,16 we recoded the six possible response choices into three categories: i) ‘Most of the time’ (response choices 1, 2, and 3); ii) ‘Some of the time’ (response choices 4 and 5); and iii) ‘None of the time’ (response choice 6). In Singapore, the English and Chinese versions of the SF-36 have been cross-language validated in population surveys14,18 and they have also been used in a previous study of patients with anxiety disorders.19 VOL. 95-B, No. 11, NOVEMBER 2013

1491

Functional ability was measured by three items from the Knee Society Clinical Rating System20: walking tolerance, stairs ability, and walking aid used. We recoded these three variables due to low prevalence of some response categories. Specifically, for walking tolerance, the possible six response choices were recoded into four categories: i) ‘Unable/ housebound’ (response choices 1 and 2); ii) ‘one block/ bus-stop’ (response choice 3); iii) ‘one to four blocks/busstops’ (response choices 4 and 5); and iv) ‘Unlimited’ (response choice 6). For stairs ability, the possible five response choices were recoded into three categories: i) ‘Unable’ (response choice 1); ii) ‘Full rail support needed’ (response choices 2 and 3); and iii) ‘Partial or no rail support’ (response choices 4 and 5). For walking aid used, the possible four response choices were recoded into three categories: i) ‘None’ (response choice 1); ii) ‘Quadstick or cane’ (response choice 4); and iii) ‘Two-canes or crutches/ walker’ (response choices 2 and 3). We assessed the discharge plan and social support by asking patients: ‘What is your preferred discharge destination (home or rehabilitation facility)?’ and ‘Will there be someone to care for you after your operation (yes or no)?’ These two questions originated from an inventory developed by Oldmeadow, McBurney and Robertson21 to assess the risk for extended inpatient rehabilitation after hip or knee replacement. Other candidate predictors included previous history of knee replacement (yes or no), type of surgery (unicompartmental versus total knee replacement, unilateral versus bilateral knee replacement) and surgeon volume (high versus low): we defined high-volume as > 100 operations per year22 as this definition also identified the adult reconstruction specialists (n = 8) in our study. Our outcome of interest was hospital LOS, and this refers to the number of nights spent in the hospital from the day of surgery. Statistical analysis. We used descriptive statistics to characterise the study sample: means with standard deviations (SDs) and medians with interquartile ranges (IQR) for continuous variables and frequencies with percentages for categorical variables. Less than 1% of patients (n = 12) had missing data for any given characteristic and we have no missing LOS data. Given the low rate of missing data,23 we used a flexible additive regression model to singly impute missing values,24,25 incorporating into the imputation procedure all available variables. In order to complement the multivariable regression analyses, patients with and without a long LOS (dichotomised at > five days based on the third quartile value) were compared by various demographic and clinical characteristics using Wilcoxon rank sum test for continuous variables and Pearson chi-squared test for categorical variables. To develop the prediction model, we used multivariable linear regression and included all 17 candidate variables as listed in Table I. To allow for possible nonlinear associations between the continuous predictors and LOS,26 age was modelled using a three-knot restricted cubic

1492

P-H. ONG, Y-H. PUA

Table I. Demographics and clinical characteristics of patients by length of stay (IQR, interquartile range) Length of stay *

Characteristic

Combined

≤ 5 days

> 5 days

p-value

Patients (n) Mean (SD) length of stay (days) [median; IQR] Mean (SD) age (yrs) [median; IQR] Male (n, %) Mean (SD) BMI (kg/m2) [median; IQR] Mean (SD) SF-36 bodily pain [median; IQR]

1609 5 (2) [4; 4 to 5] 66 (8) [66; 61 to 72] 323 (20.1) 28 (4) [27; 25 to 30] 35 (19) [32; 22 to 51]

1258 4 (1) [4; 3 to 4] 66 (7) [65; 61 to 71] 264 (21.0) 28 (4) [27; 25 to 30] 36 (19) [32; 22 to 51]

351 8 (2) [7; 6 to 8] 69 (8) [69; 64 to 75] 59 (16.8) 27 (4) [27; 24 to 30] 32 (18) [31; 22 to 42]

< 0.001† < 0.001† < 0.084‡ < 0.4† < 0.001 †

Comorbidities (n, %) 0 1 to 3 ≥4

132 (8.2) 1152 (71.6) 325 (20.2)

110 (8.7) 920 (73.1) 228 (18.1)

22 (6.3) 232 (66.1) 97 (27.6)

Mean (SD) ROM (°) [median; IQR] Knee flexion Knee extension

119 (19) [122; 110 to 133] 120 (17) [123; 110 to 133] 116 (23) [120; 108 to 130] < 0.003† 7 (9) [5; 1 to 10] 7 (8) [5; 1 to 10] 7 (9) [5; 2 to 10] < 0.925†

< 0.003§

< 0.001§

Down and depressed (n, %) Most of the time Some of the time None of the time

222 (13.8) 596 (37.0) 791 (49.2)

154 (12.2) 452 (35.9) 652 (51.8)

68 (19.4) 144 (41.0) 139 (39.6)

Walking ability (n, %) Unable/housebound 1 bus-stop 1 to 4 bus-stops Unlimited

302 (18.8) 526 (32.7) 602 (37.4) 179 (11.1)

205 (16.3) 409 (32.5) 485 (38.6) 159 (12.6)

97 (27.6) 117 (33.3) 117 (33.3) 20 (5.7)

Walking aids (n, %) None Quadstick or cane 2 canes or frame

1030 (64.0) 528 (32.8) 51 (3.2)

862 (68.5) 368 (29.3) 28 (2.2)

168 (47.9) 160 (45.6) 23 (6.6)

Stairs ability (n, %) Unable Full rail support Partial or no rail support

112 (7.0) 1427 (88.7) 70 (4.4)

68 (5.4) 1128 (89.7) 62 (4.9)

44 (12.5) 299 (85.2) 8 (2.3)

< 0.001§

Previous knee surgery (n, %) Bilateral surgery (n, %) Surgery type (TKR) (n, %) Surgeon volume (High) (n, %) Carer (Available) (n, %) Patient expectation (Step-down care) (n, %)

412 (25.6) 68 (4.3) 1463 (90.9) 1222 (75.9) 1325 (82.3) 148 (9.2)

327 (26.0) 30 (2.4) 1122 (89.2) 1007 (80.0) 1083 (86.1) 32 (2.5)

85 (24.2) 38 (10.8) 341 (97.2) 215 (61.3) 242 (68.9) 116 (33.0)

0.5‡ < 0.001‡ < 0.001‡ < 0.001‡ < 0.001‡ < 0.00‡

< 0.001§

< 0.001§

* BMI, body mass index; SF-36, Short-Form 36; ROM, range of movement; TKR, total knee replacement † Wilcoxon rank-sum test ‡ Pearson chi-squared test § Proportional odds likelihood ratio test

spline.25 To parse the model, a backward stepwise selection procedure was used and we retained variables with a liberal p-value ≤ 0.20 to ensure that potentially important predictors were selected.27 Because the distribution of LOS was right-skewed, it was log-transformed before performing the regression analyses. After ascertaining that the log-scale residuals of the final regression model were homoscedastic,28 we used the Duan smearing estimator29 to adjust for the bias arising from re-transforming the log-LOS values to its original scale (days). For ease of use in the clinical setting,

we presented the prediction model as a nomogram that allows one to estimate the LOS for individual patients. We used two measures to assess the model’s performance. First, predictive performance was measured by the R2 coefficient, which is the percentage variance explained by the model. Because a prediction model can always be expected to perform better (optimistically) in the original derivation sample than in new but similar samples, we used a bootstrap internal validation technique to correct (shrink) the R2 value for ‘optimism’.25 Second, model calibration THE BONE & JOINT JOURNAL

A PREDICTION MODEL FOR LENGTH OF STAY AFTER TOTAL AND UNICOMPARTMENTAL KNEE REPLACEMENT

1493

Table II. Statistical analysis of variance for prediction model Variable*

F-value

Degrees of freedom

p-value

Age Nonlinear age Number of comorbidities Knee flexion ROM Down and depressed Walking aid Surgery type (TKR) Bilateral surgery Surgeon volume Carer Patient expectation

13.21 4.99 4.90 7.55 9.34 10.22 50.66 69.96 72.35 7.76 264.77

2 1 2 1 2 2 1 1 1 1 1

< 0.001 0.0257 0.0075 0.0061 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 0.0054 < 0.001

59.10

14

< 0.001

Total

* ROM, range of movement; TKR, total knee replacement

was assessed graphically using a calibration plot of observed LOS against predicted LOS. Statistical analyses were carried out using R software v2.15.0 (R Foundation, Vienna, Austria), using the rms package.24 Statistical significance was determined at the two-sided p-value < 0.05.

Results Table I shows the sociodemographic and clinical characteristics of the sample. The patients were predominantly female (80%) and were on average moderately overweight (mean 27.7 kg/m2 (SD 4.4)). The median LOS for the overall sample was four days (IQR 4 to 5). There were statistically, but not always clinically, significant differences between patients with and without a long LOS (> five days). Patients with a long LOS were somewhat older, had a lower pre-operative range of knee flexion, and had somewhat greater knee pain than patients with a short LOS. Patients in the long LOS group were also more likely to undergo a total or bilateral knee replacement, to be operated on by low-volume surgeons, and to report frequent feelings of being down and depressed. In terms of functional ability, patients with a long LOS had lower self-reported walking and stair climbing abilities and they were more likely to use a walking aid with a larger baseof-support. Finally, patients with a long LOS were less likely to report having a carer at home and they were more likely to expect step-down care post-surgery. (Table I). Prediction model. Table II shows the results of multivariable regression modelling. Statistically significant predictors of longer LOS included older age, greater number of comorbidities, a lower range of knee flexion pre-operatively, frequent feelings of being down and depressed, greater dependence of walking aids for support, bilateral knee replacement, having surgery by a low-volume surgeon, absence of a carer at home, and an expectation of receiving step-down care. Figure 1 shows the nomogram constructed based on these ten variables, which allows the user to predict the LOS for individual patients. The R2 value VOL. 95-B, No. 11, NOVEMBER 2013

of the prediction model was 0.34 and, after correction for optimism, 0.32, indicating minimal model optimism (< 2%). Figure 2 shows the calibration plot of observed and predicted LOS following knee replacement. The calibration plot showed reasonable agreement between the observed and predicted LOS values (r = 0.58) and 95% of their differences lie approximately between –3 and +3 days.

Discussion We have developed and internally validated a prediction model for LOS in 1609 patients who have undergone total or unicompartmental knee replacement. Although previous studies have examined the factors associated with a prolonged LOS, few have integrated these factors into a single model to predict absolute LOS. Thus, we believe our study is the largest and most contemporary study to date of absolute LOS prediction in knee replacement. Our study evaluates a comprehensive range of patient predictors, and most variables in the final model – namely, age,2-4,6,9,10 number of comorbidities,3,5-8 range of knee movement,9 walking aids used,2,3,9 surgery type2 and surgeon volume (a proxy measure of surgeon specialist in our study)2 – have empirical support from previous studies investigating LOS, thereby providing some confidence for the model’s applicability. In our model, having a carer at home was predictive of a shorter LOS while patients’ expectations about their discharge destination contributed the greatest number of nomogram points (Fig. 1). These predictors are part of a validated inventory21 to assess the risk for extended inpatient rehabilitation after knee replacement in an Australian healthcare setting. In the context of our study, patients who were discharged to stepdown facilities usually required extended inpatient rehabilitation and it has been our experience and that of others 30 that social and organisational factors could also exert a considerable influence on increasing their LOS in the acute hospital. For these reasons, our findings are both intuitive and unsurprising.

1494

P-H. ONG, Y-H. PUA

0

10

20

30

40

50

60

70

80

90

100

Points Age 50

70

75

80

85

90

1 to 3

No. of comorbidities Knee flexion ROM

≥4

None

160

140

120

100

80

60

40

20

0

Some of the time

Down and depressed

None of the time

Most of the time Quadstick or cane

Walking aid None

2 canes or frame TKR

Surgery type UKR

Yes

Bilateral surgery No Low

Surgeon volume High Absent

Carer Present

Step−down care

Patient expectation Home

Total points

0

50

100

150

200

250

300

350

Predicted LOS (days) 3

3.5

4

4.5

5

5.5

6

6.5

7

7.5

8

8.5

9

9.5

−1

−0.5

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

10 10.5 11 11.5 12 12.5

Reference to median LOS 6

6.

7

7.5

8 8.5

Fig. 1 Nomogram of multivariable prediction model for length of stay (LOS) after knee replacement. To predict the LOS, draw a vertical line from each predictor to the top line, labelled ‘Points’ to calculate points associated with each predictor. The sum of these points is then marked on the line labelled ‘Total points’. Drop a vertical line from there to determine the predicted LOS or the predicted LOS with reference to the median LOS (4 days) (ROM, range of movement)

We found that 14% of our patients reported feeling down and depressed ‘most of the time’ (Table I), a prevalence rate of depression which was comparable with that reported in the Osteoarthritis Initiative cohort (11.8%).31 Negative psychological symptoms are not trivial as a sizeable body of literature suggests that high pre-operative levels of depression and internalising/ catastrophising were associated with lesser improvements in pain and physical functioning after knee replacement.31-35 Despite its prevalence and impact, the knee replacement literature is surprisingly light in reporting the influence of pre-operative psychological symptoms on LOS, with only one small study (n = 43)33 demonstrating that high pre-operative levels of pain and catastrophising were predictive of a prolonged LOS. Against this background, our results expand on Witvrouw et al’s33 findings that indicate that pre-operative depressive symptoms were also predictive of LOS, although we acknowledge two caveats. First, we used a single SF-36 item16 to assess depression and although previous work has shown that even a single question may be used to identify patients with depression,17,36 future studies may benefit from a more comprehensive assessment of depression using

SF-36 and other depression scale. Second, we did not assess the level of pain catastrophising in our patients, and although these two measurements are interrelated, they are nevertheless distinct constructs.37 Accordingly, it may be necessary for future investigations to examine a comprehensive body of psychological instruments and to compare all measures simultaneously using a multivariable approach. Our study has potential implications for clinical practice, research, and hospital management policy. Our prediction model uses pre-operative registry data that are easily or routinely obtained in patients prior to undergoing knee replacement which, in turn, facilitate its integration in clinical practice as a decision-making support aid. Our model could be used to identify high-risk patients (e.g. older patients with severe functional limitations or suspected depression) for enrolment in targeted pre-operative38,39 and acute post-operative40,41 interventions to modify their risk. Also, our model may be useful for managing surgery schedules and hospital bed allocations, so improving the costeffectiveness of care in knee replacement. Early knowledge of an accurate expected LOS can also help shape patient’s THE BONE & JOINT JOURNAL

A PREDICTION MODEL FOR LENGTH OF STAY AFTER TOTAL AND UNICOMPARTMENTAL KNEE REPLACEMENT

1495

transferring to step-down care. In this group of patients, while our institution provides weekend physiotherapy services for all patients,40 their LOS will still be determined by the operating hours of the accepting facility. Although not perfect, we should mention that our prediction model is still better than one that assumes a LOS of four days (the sample median) across all patients: such a model not only has a wider prediction interval (four vs three days), but also will under-estimate the LOS of patients (38% of our sample) whose LOS was five days or longer (total underestimation in days: 1642 vs 914 days using our model). In conclusion, we have developed a prediction model for absolute LOS in knee replacement based on ten easy-todetermine preoperative variables from our registry database. Given that knee registries are increasingly available in many countries, we hope that our findings will stimulate further study, first to externally validate our model and then to determine whether application of the prediction model would lead to better clinical outcomes and costeffectiveness of care in knee replacement. Fig. 2 Calibration plot of observed length of stay (LOS) against predicted LOS. Regression (solid) line represents the mean observed LOS value for each predicted LOS value. The shaded region is the 95% prediction interval for an individual predicted LOS value. Dotted line is the line of equality (y = x).

expectations and facilitates discharge planning. To help translate our findings to practice and policies, these purported benefits and applications of our prediction model should be formally evaluated in cluster randomised studies and cost-effectiveness analyses. Our study has limitations. First, our single-institution study raises questions about the general applicability of our prediction model. However, our sample is nationally representative because our institution performed more than half (60%) of all knee replacements in Singapore42 and our patients were similar to those from another local institution in terms of key observable demographic and physical characteristics.43 Nevertheless, we acknowledge the need to externally validate and update our model in new patient cohorts from our and other institutions.44 Second, due to differences in healthcare settings and practice, it is unlikely that a prediction model developed in one country would be directly applicable in another. Nevertheless, because our prediction model was developed using rigorous statistical methods, our paper illustrates the steps involved in developing a prediction model and our findings should serve as useful baseline data in the development or modification of similar models in other countries or healthcare settings. Finally, the adjusted R2 value of our final model was 0.32, indicating that a notable amount of variation of LOS was unexplained and attesting to the multifactorial influences on LOS. We acknowledge that the day of surgery could potentially influence LOS particularly for those patients VOL. 95-B, No. 11, NOVEMBER 2013

Supplementary material Two tables, detailing i) post-operative complications in excluded patients and ii) a detailed description of the model for prediction of length of stay, are available with the electronic version of this article on our website www.bjj.boneandjoint.org.uk. The authors would like to thank Ms. B. Y. Tan, the head of the Department of Physiotherapy, Singapore General Hospital, for supporting this study. They also thank Ms. H. C. Chong and Mr. W. Yeo from the Orthopaedic Diagnostic Centre, Singapore General Hospital, for their assistance. Finally, they would like to thank Mr. J. S. J Wong for his research assistance and the orthopaedic surgeons from the Singapore General Hospital for allowing access to their patients. No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article. This article was primary edited by D. Rowley and first-proof edited by G. Scott.

References 1. Kurtz S, Ong K, Lau E, Mowat F, Halpern M. Projections of primary and revision hip and knee arthroplasty in the United States from 2005 to 2030. J Bone Joint Surg [Am] 2007;89-A:780–785. 2. 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;90B:1435–1440. 3. Husted H, Holm G, Jacobsen S. Predictors of length of stay and patient satisfaction after hip and knee replacement surgery: fast-track experience in 712 patients. Acta Orthop 2008;79:168–173. 4. 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;13:186– 190. 5. Kimmel LA, Oldmeadow LB, Sage C, Liew SM, Holland AE. A designated three day elective orthopaedic surgery unit: first year's results for hip and knee replacement patients. Int J Orthop Trauma Nurs 2011;15:29–34. 6. MacDonald V, Ottem P, Wasdell M, Spiwak R. Predictors of prolonged hospital stays following hip and knee arthroplasty. Int J Orthop Trauma Nurs 2010;14:198– 205. 7. Epps CD. Length of stay, discharge disposition, and hospital charge predictors. AORN J 2004;79-5:975–997. 8. Wasielewski RC, Weed H, Prezioso C, Nicholson C, Puri RD. Patient comorbidity: relationship to outcomes of total knee arthroplasty. Clin Orthop Relat Res 1998;356:85–92. 9. 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;36:1601–1608.

1496

P-H. ONG, Y-H. PUA

10. Rissanen P, Aro S, Paavolainen P. Hospital- and patient-related characteristics determining length of hospital stay for hip and knee replacements. Int J Technol Assess Health Care 1996;12:325–335. 11. Lin JJ, Kaplan RJ. Multivariate analysis of the factors affecting duration of acute inpatient rehabilitation after hip and knee arthroplasty. Am J Phys Med Rehabil 2004;83:344–352. 12. Dauty M, Smitt X, Menu P, Dubois C. Which factors affect the duration of inpatient rehabilitation after total knee arthroplasty in the absence of complications? Ann Phys Rehabil Med 2009;52:234-45 (in English, French).:. 13. Sangha O, Stucki G, Liang MH, Fossel AH, Katz JN. T The Self-Administered Comorbidity Questionnaire: a new method to assess comorbidity for clinical and health services research. Arthritis Care Res 2003;49:156–163. 14. Thumboo J, Fong KY, Machin D, et al. A community-based study of scaling assumptions and construct validity of the English (UK) and Chinese (HK) SF-36 in Singapore. Qual Life Res 2001;10:175–188. 15. Ware JE Jr, Sherbourne CD. The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med Care 1992;30:473–483. 16. Paz SH, Globe DR, Wu J, Azen SP, Varma R. Relationship between self-reported depression and self-reported visual function in Latinos. Arch Ophthalmol 2003;121:1021–1027. 17. Lopes AA, Bragg J, Young E, et al. Depression as a predictor of mortality and hospitalization among hemodialysis patients in the United States and Europe. Kidney Int 2002;62:199–207. 18. Thumboo J, Fong KY, Chan SP, et al. The equivalence of English and Chinese SF36 versions in bilingual Singapore Chinese. Qual Life Res 2002;11:495–503. 19. Luo N, Fones CS, Thumboo J, Li SC. Factors influencing health-related quality of life of Asians with anxiety disorders in Singapore. Qual Life Res 2004;13:557–565. 20. Insall JN, Dorr LD, Scott RD, Scott WN. Rationale of the Knee Society clinical rating system. Clin Orthop Relat Res 1989;248:13–14. 21. Oldmeadow LB, McBurney H, Robertson VJ. Predicting risk of extended inpatient rehabilitation after hip or knee arthroplasty. J Arthroplasty 2003;18:775–779. 22. Lavernia CJ, Guzman JF. Relationship of surgical volume to short-term mortality, morbidity, and hospital charges in arthroplasty. J Arthroplasty 1995;10:133–140. 23. Barzi F, Woodward M. Imputations of missing values in practice: results from imputations of serum cholesterol in 28 cohort studies. Am J Epidemiol 2004;160:34–45. 24. Harrell FE Jr. rms: Regression Modeling Strategies. R package version 3.5-0. 2012. http://CRAN.R-project.org/package=rms (date last accessed 4 June 2013). 25. Harrell FE Jr. Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis. New York: Springer, 2001. 26. Desquilbet L, Mariotti F. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med 2010;29:1037–1057. 27. Steyerberg EW, Eijkemans MJ, Harrell FE Jr, Habbema JD. Prognostic modeling with logistic regression analysis: in search of a sensible strategy in small data sets. Med Decis Making 2001;21:45–56.

28. Manning WG, Mullahy J. Estimating log models: to transform or not to transform? J Health Econ 2001;20:461–494. 29. Duan N. Smearing estimate: a nonparametric retransformation method. J Am Statist Assoc 1983;78:605–610. 30. Boxall A, Sayers A, Caplan GA. A cohort study of 7 day a week physiotherapy on an acute orthopaedic ward. J Orthop Nurs 2004;8:96–102. 31. Riddle DL, Kong X, Fitzgerald GK. Psychological health impact on 2-year changes in pain and function in persons with knee pain: data from the Osteoarthritis Initiative. Osteoarthritis Cartilage 2011;19:1095–1101. 32. Brander V, Gondek S, Martin E, Stulberg SD. Pain and depression influence outcome 5 years after knee replacement surgery. Clin Orthop Relat Res 2007;464:21–26. 33. Witvrouw E, Pattyn E, Almqvist KF, et al. Catastrophic thinking about pain as a predictor of length of hospital stay after total knee arthroplasty: a prospective study. Knee Surg Sports Traumatol Arthrosc 2009;17:1189–1194. 34. Lingard EA, Riddle DL. Impact of psychological distress on pain and function following knee arthroplasty. J Bone Joint Surg [Am] 2007;89-A:1161–1169. 35. Sullivan M, Tanzer M, Stanish W, et al. Psychological determinants of problematic outcomes following Total Knee Arthroplasty. Pain 2009;143:123–129. 36. Reme SE, Eriksen HR. Is one question enough to screen for depression? Scand J Public Health 2010;38:618–624. 37. Edwards RR, Cahalan C, Mensing G, Smith M, Haythornthwaite JA. Pain, catastrophizing, and depression in the rheumatic diseases. Nat Rev Rheumatol 2011;7:216–224. 38. Topp R, Swank AM, Quesada PM, Nyland J, Malkani A. The effect of prehabilitation exercise on strength and functioning after total knee arthroplasty. PM&R 2009;1:729–735. 39. Somers TJ, Blumenthal JA, Guilak F, et al. Pain coping skills training and lifestyle behavioral weight management in patients with knee osteoarthritis: a randomized controlled study. Pain 2012;153:1199–1209. 40. Pua YH, Ong PH, Chong HC, Lo NN. Sunday physiotherapy reduces inpatient stay in knee arthroplasty: a retrospective cohort study. Arch Phys Med Rehabil 2011;92:880–885. 41. Oldmeadow LB, McBurney H, Robertson VJ, Kimmel L, Elliott B. Targeted postoperative care improves discharge outcome after hip or knee arthroplasty. Arch Phys Med Rehabil 2004;85:1424–1427. 42. No authors listed. Ministry of Health. Knee replacement surgery, 2001. http:// www.moh.gov.sg/content/moh_web/home/costs_and_financing/HospitalBillSize/ knee_replacementsurgery.html (date last accessed 5 June 2013). 43. Xu GG, Sathappan SS, Jaipaul J, Chan SP, Lai CH. A review of clinical pathway data of 1,663 total knee arthroplasties in a tertiary institution in Singapore. Ann Acad Med Singapore 2008;37:924–928. 44. Moons KG, Kengne AP, Grobbee DE, et al. Risk prediction models: II. External validation, model updating, and impact assessment. Heart 2012;98:691–698.

THE BONE & JOINT JOURNAL

A prediction model for length of stay after total and unicompartmental knee replacement.

Early and accurate prediction of hospital length-of-stay (LOS) in patients undergoing knee replacement is important for economic and operational reaso...
496KB Sizes 0 Downloads 0 Views