Anesthesiologists administered spinal anesthesia with hyperbaric local anesthetics (Marcaine Spinal 0.5% Heavy, Astra Zeneca, Sweden) and fentanyl 10-15 μg for unilateral TKA. The objective block level was T10, and after confirming the appropriateness of the block level, sedation was performed during the surgery using propofol or dexmedetomidine. Postoperative pain management was performed using fentanylbased intravenous (IV) patient-controlled analgesia (PCA). The regimen, based on the patient’s weight and underlying diseases, was as follows: fentanyl 8-15 μg ml⁻¹, 1 ml bolus dose without background infusion, and a 10-min lockout time. IV PCA (100 ml total) was mostly used on postoperative day (POD) 0-1, and additional opioid analgesics (e.g., morphine, fentanyl, oxycodone) were used after IV PCA, as per the patient’s request or the surgeon’s decision. For adjuvant pain control, either periarticular injection was given by the surgeon during surgery, or femoral nerve block was administered by the anesthesiologist. While 0.375% ropivacaine with morphine was used for intraoperative periarticular injection, 0.2% ropivacaine was infused through a catheter for femoral nerve block, at a rate of 4 ml per hour, from the start of the surgery until 48 h after the operation. To avoid the effects of local anesthetic toxicity, the attending physician first aspirated through the needle before injection to avoid blood injection, and injected a small test dose (1-2 ml) to check for immediate toxicity before injecting the full dose.

The location and intensity of pain at all sites experienced by the patients 1-2 days prior to TKA surgery are routinely recorded at our institution. The pain can be classified as a mixed-type pain, which involves somatic and neuropathic pain. The information on location of pain includes all the body sites at which the patient experienced pain at admission. Pain intensity was represented using the numeric rating scale (NRS) score, ranging from 0 to 10. The patients were divided into two groups (based on the pain location as only knee pain and knee pain with external to knee pain) according to the preoperative pain reports. Postoperative pain intensity was also checked using the NRS scores during rest and during movement, and all checks were regularly performed once every 4-6 h by registered nurses.

Data collected for the study included demographic information (age, sex, and body mass index), American Society of Anesthesiologists physical status, Charlson comorbidity index [12], intraoperative sedative uses (propofol or dexmedetomidine), preoperative analgesics use (opioid analgesics and non-opioid analgesics) and postoperative pain outcome for POD 0-3 (morphine equivalent consumption, NRS pain scores, and adjuvant analgesics use [acetaminophen and ketorolac]). Preoperative non-opioid analgesics did not include anticonvulsants or antidepressants, and generally, all preoperative regular medications were re-started after surgery at POD 0. The postoperative NRS pain scores during rest and movement were represented using an average daily value of the multiple assessments made on each day, and morphine equivalent consumption was derived by converting all intravenous and oral opioid analgesics used by the patients into their morphine equivalents (mg), according to the standard conversion ratio [13] (Table 1). All medical records were collected by medical record technicians who were blinded to the purpose of the study. The main researchers of the study were blinded to the patient characteristics until after the statistical analyses had been completed.

The baseline characteristics of the patients were represented using mean with standard deviation (SD) and number with percentage. In order to assess how each variable affects the morphine equivalent consumption for POD 0-3, we performed a generalized linear regression analysis. In the univariable generalized linear model (GLM), the covariates with P < 0.2 for our two main independent variables (preoperative pain and analgesics use) were then selected for further analysis using a multivariable GLM.

Next, we performed 10 multivariable GLMs according to the preoperative pain site for 10 dependent variables: postoperative NRS 0, 1, 2, and 3, and adjuvant analgesic consumption (acetaminophen and ketorolac) after TKA. The backward variable selection method with an elimination criterion of was used to fit the final multivariable model for these 10 dependent variables. For the GLMs, gamma distribution was assumed and the log link function was used for all dependent variables (morphine equivalent consumption, postoperative NRS 0, 1, 2, and 3, and adjuvant analgesic consumption after TKA). The results of GLM were presented as exponentiated regression coefficient (Exp. Coef) with 95% confidence interval (CI). IBM SPSS version 24.0 (IBM Corp., USA) was used for all analyses, with the statistical significance set at P < 0.05.

Table 3 shows the results of the univariable GLM for the morphine equivalent consumption on POD 0-3. The results of the multivariable GLM using the variables with P < 0.2 from the univariable GLM are presented in Table 4. The results of the final multivariable GLM are shown in Table 4. Morphine equivalent consumption on POD 0-3 was significantly 5% higher in the group with preoperative knee pain with external to knee pain, compared to the patients who reported preoperative knee pain only (Exp Coef: 1.05; 95% CI, 1.02 to 1.09; P = 0.004). Additionally, the morphine equivalent consumption of patients in the non-opioid analgesic use group and opioid analgesic use group was not significantly different from that of patients in the “no preoperative regular analgesics use” group (P > 0.05). Among other variables, femoral nerve block and intraoperative propofol use were associated, respectively, with a 11% (Exp. Coef: 0.89; 95% CI: 0.87 to 0.92; P < 0.001) and 4% (Exp. Coef: 0.96; 95% CI: 0.93 to 0.99; P = 0.004) lower morphine equivalent consumption during POD 0-3.

Table 5 shows the results of the multivariable GLM performed using the NRS pain score on PODs 0, 1, 2, and 3, and adjuvant analgesic use (acetaminophen and ketorolac) as dependent variables, along with backward variable selection using an elimination criterion of 0.05. No significant association was found between the site of pain (knee only vs. knee pain with external to knee pain) and these factors (P > 0.05).