Age showed weak inverse correlation with duration of surgery (r = −0.14, p = 0.012). Surgery time diminished with increasing patient age. Patient height did not correlate with duration of surgery (r = 0.02, p = 0.7). Weight correlated positively with duration (r = 0.3, p < 0.001) as did body mass index (BMI; r = 0.3, p < 0.001). The heavier patient has a longer operative time.
The results for the nominal variables are shown in Table . With regard to the operated side, having both sides done differs from either of the single sides (p < 0.001 for each). Thus, having both sides done increases the operation time. Those with a diagnosis of osteoarthritis (p = 0.01) or avascular necrosis (p = 0.049) have shorter operation times than posttraumatic subjects. The frequency of other diagnoses is too low to show statistically significant effects. Patients with no other comorbidities had operation times of 93±28 min (n = 236), whereas patients with other comorbidities had operation times of 127 ± 45 min (n = 104; p < 0.001). The correlation between the duration of surgery and the total number of comorbidities was statistically significant (r = 0.2, p < 0.001; n = 340).
Association of surgical time with nominal data
A sum of risks was calculated from five modalities considered as a risk factor: having a revision, having both sides done, the existence of any comorbidities, morbid obesity, and any prior open surgery. This produced a discrete scale from 0 to 5. The diagnoses were also reclassified and combined to three groups: prior sepsis or avascular necrosis, osteoarthritis or rheumatoid arthritis, and all other diagnoses. This will be referred to as the revised DX. The actual observed scores ranged from 0 to 3. No subjects had scores higher than 3. The results are shown in Table . The duration of the operation increased as the number of risks increased, and this linear trend is statistically significant (p < 0.001). However, variances are large.
Duration of surgical operating time and sum of risks
In a stepwise linear regression, the risks and BMI together explain 24% of the variance. The weight cubed is the best fit of several weight functions applied to the data. In a stepwise linear regression, the risks and weight cubed together explain 25% of the variance. The largest amount of variance (31%) is explained by BMI, side (unilateral vs. bilateral), whether the patient had prior operation or not, whether the patient is morbidly obese or not, and primary/revised TKR. This includes the weight twice both as BMI and in the obese function. Removing five outliers increases the explained variance to 32%, with predictors risk summation, weight, morbid obesity, prior open surgery, and the revised DX.
Patients without infections (n
= 236) had surgery durations of 94 ± 28 min. Patients with infection (n
= 104) had durations of 127 ± 45 min (p
< 0.001). This duration (127 min) can be interpreted as the critical duration of surgery in the perspective of increased infection risk. The predictors of infection have been discussed in a paper by Peersman et al. [7
]. The duration of surgery is a predictor of infection, together with side of operation, morbid obesity, reclassified diagnosis, and BMI. If the cutoff value is set at 0.25, 72% of the cases are correctly assigned, 168 of 236 uninfected, and 73 of 101 infected. Adding other variables such as surgeon did not improve the prediction.
Nonlinear modeling was applied to the data, but prediction of either duration or infection was inferior to the models described above.
None of the models predict 50% or more of the variance. It is surprising that in predicting the infection rate, duration is joined by variables such as morbid obesity and diagnosis, which are independently associated with duration. This suggests a nonlinear association and the possibility of an underlying factor not measured in the current study.