The Chitranjan Ranawat Award

Abstract

Total knee replacement effectively relieves arthritis pain but improvement in physical function varies. A clearer understanding of the patient attributes associated with differing levels of functional gain after TKR is critical to surgical decision making. We reviewed 8050 primary, unilateral TKR patients enrolled in a prospective registry between 2000 and 2005 who had complete data. We evaluated associations between 12-month function (SF12/PCS) and preoperative gender, age, BMI, emotional health (MCS), knee diagnosis, quadriceps strength, and physical function (PCS). More than 98% of patients reported pain relief (KS pain score). At 12 months, mean PCS gain was 13.6 points, but the distribution was bimodal. The mean gain in PCS in the 63% of patients with greater improvement was 21 (SD = 7), and 4.1 (SD = 7) in the remaining 37%. Increased likelihood of poor functional gain was associated with older age, body mass index (BMI) over 40, lower MCS, and poor quadriceps strength. While two-thirds of patients reported functional gain well above national average at 12 months post-TKR, 37% reported limited functional improvement. Further understanding of the patient attributes associated with limited improvement will guide the design of innovative strategies to improve functional outcomes.

Level of Evidence: Level II, prognostic study. See the Guidelines for Authors for a complete description of levels of evidence.

Appendices

Appendix 1

Comparison of patient attributes in PORT study, Zimmer total, Zimmer subset with complete data, and single site registry

Pre-TKR Patient attributes	Zimmer subset (Research Cohort)*	Zimmer total	PORT study†	Central NY: Single site
Years of enrollment	2000–2004	2000–2004	1992–1993	1998–2000
# Surgeons	135	171	48	2
# Patients eligible for inclusion	unknown	unknown	563	172
# Patients included	8050	17,270	291	172
% Patients included	unknown	unknown	52	100
% Patients with followup data*	100 (1 year)	51 (1 year)	92 (2 years)	100 (1 year)
Patients Included
Osteoarthritis diagnosis (%)	95	94	100	100
Knee pain past 6 M (%)	100	100	100	100
Primary TKR (%)	100	100	100	100
Potential Prognostic Variables
Age (mean years)	67.8	67.6	70.2	67.3
Male %	34	35	37	34
Female %	66	65	63	66
Region of US (# states)	31	36	Indiana	Central NY
BMI (mean)	32	32	30	—
SF12 PCS preoperative	30.3	30.2	27.4	30.4
SF12 MCS preoperative	52.6	53.6	52.5	52.5
KSS- Pain preoperative	37.0	35.6	34.7	—
KSS- Function preoperative	47.0	47.7	41.2	—
Outcome Variables
SF12 PCS 12 month	43.8	42.6	38.1	40.8
Mean PCS change (pre to 12 mo)	13.6	14.1	10.5	10.4
KSS- Pain 6 month	80.0	76.8	62.8	—
KSS- Function 6 month	67.3	66.1	62.5	—
Surgeon and hospital factors
% patients treated by surgeons with annual volume > 20/year	78%	73%	77%	100%
Surgeon volume > 20/year in database (%)	30.4%	22.5%	—	100%
% patients treated in hospitals with annual volume < 50	40%	39%	—	0
Hospital < 50 TKR/year (%)	41% (n = 92)	45% (n = 144)	20% (n = 5)	0

1. ^*Defined as cases with complete baseline and 12-month followup data.^†Sources: Tierney, 1994 [21]; Heck, 1998 [12].

Appendix 2

Mixture Model Detail

The mixture model can be written as: $y_{i} = {\mathbf{x}}_{i} {\varvec{\upbeta}} + \pi _{1i} e_{1} + \pi _{2i} e_{2} + \varepsilon _{i}$, where $y_{i}$ is 12-month PCS change of subject $i$, ${\mathbf{x}}_{i}$ `is the vector of baseline characteristics of the subject, ${\varvec{\upbeta}}$ are the coefficients, $\varepsilon _{i} \sim N\left( {0,\sigma ^{2} } \right)$; $\pi _{1i} = {{\exp \left( {{\mathbf{z}}_{i} {\varvec{\upalpha}}} \right)} \mathord{\left/ {\vphantom {{\exp \left( {{\mathbf{z}}_{i} {\varvec{\upalpha}}} \right)} {\left( {1 + \exp \left( {{\mathbf{z}}_{i} {\varvec{\upalpha}}} \right)} \right)}}} \right. \kern-\nulldelimiterspace} {\left( {1 + \exp \left( {{\mathbf{z}}_{i} {\varvec{\upalpha}}} \right)} \right)}}$ and $\pi _{2i} = 1 - \pi _{1i}$ are the mixing probabilities for latent low- and higher-responder groups, ${\mathbf{z}}_{i}$ is the vector of the $i$-th subject’s characteristics that predict his/her probability of belonging to the low responder group, and $e_{1}$ and $e_{2}$ are the two discrete random variables corresponding to the low- and high-responder groups. As part of the model, we used a logit function to model the probability of a patient outcome falling in the low functional improvement group (versus high functional gain). A patient was classified into the low functional response group if $\pi _{1i} > \pi _{2i}$.