Study Design and Participants
All subjects in this retrospective study provided written informed consent for their participation, and the process was approved by the research ethics committee of the Department of Pain Management at the First Affiliated Hospital, School of Medicine, Zhejiang University (IIT2021116). The subjects comprised patients with ZAP who underwent PRF and met the inclusion and exclusion criteria for this study, conducted between January 2017 and October 2020.
The inclusion criteria were as follows: (1) age > 40 years; (2) location of pain consistent with the HZ region and pain originating from HZ; (3) moderate to severe pain with numerical rating scale (NRS) score ≥ 4 before enrollment after routine treatment, such as anticonvulsants, antidepressants, opioids, or physical treatments, for 2 weeks.
The exclusion criteria were as follows: cognitive deficits or mental disorders affecting the patient’s treatment or questionnaire evaluation, serious cardio-cerebrovascular disease, hepatorenal insufficiency, infectious diseases, and other contraindications for PRF. Patients who were lost to follow-up for 3 months after PRF were also excluded from the present analysis. If the effect of PRF was not satisfactory, we adjusted the drug treatment strategy or further adopted intrathecal morphine infusion system implantation or spinal cord electrical stimulation at the appropriate time.
Study Protocol and Procedure
After admission, blood examinations were performed, including routine blood test, blood sedimentation, hypersensitive C-reactive protein, blood lipids, liver function, kidney function, immunoglobulin and complement, and tumor markers. All patients received orally administered drugs (gabapentin/pregabalin/duloxetine, nonsteroidal anti-inflammatory drugs [NSAIDs], and antiviral drugs). Dermatome corresponding to the region of HZ infection or PHN was identified if hypersensitivity and pain occurred in the corresponding HZ-affected region.
After patients were placed in an appropriate position with electrocardiographic monitoring and oxygen inhalation, they were treated with PRF (Baylis PM230 PRF generator, Baylis Medical Company, Inc., Montreal, Canada) using 18-G, straight, sharp radiofrequency cannulas with a 10-mm exposed tip, which were carefully inserted until the needle tip reached the upper edge of the intervertebral foramen. The guide methods used in the procedure were ultrasound, computed tomography (CT), radiography, ultrasound and CT, and ultrasound and radiography. After confirming the needle position with fluoroscopy guidance and sensory stimulation, the position of the needle tip was adjusted until the movement and sensation determined that the affected area was completely covered. The radiofrequency stimulation tests determined whether the HZ area was covered, as follows: (1) 50 Hz, 0.3–0.5 V test sensation, to induce pain sensation in the HZ area, and (2) 2 Hz, 0.3–0.5 V test exercise, in which pulse was induced when the nerve segment was affected. Standard PRF was conducted for 600 s at 42 ºC sequentially. All patients returned to the ward after treatments, followed by a 15-min observation period for complaints of discomfort. All PRF procedures were conducted by experienced physicians in our department.
Data Collection
The following data were collected from the medical records: sex, age, body mass index, education level, guide methods of the procedure, hypertension, diabetes, other complications (1, brain; 2, heart; 3, lung; 4, liver; 5, kidney; 6, tumor), rash site, days from HZ onset to PRF, and targeted nerve segment.
The assessment scales for pain and its concomitant symptoms were collected before and 3 months after the procedure, as follows: NRS, Brief Pain Inventory (BPI), Pain Catastrophizing Scale, Insomnia Severity Index, ID Pain, Generalized Anxiety Disorder-7, Patient Health Questionnaire-9, and Health Survey-36 (SF-36). BPI contains seven sub-items, including general activity, mood, walking ability, normal work, relations with other people, sleep, and enjoyment of life. SF-36 contains nine sub-items, including physical functioning, role–physical, bodily pain, general health, vitality, social functioning, role–emotional, mental health, and reported health transition. Doses of anticonvulsants (gabapentin mg/dL, pregabalin mg/dL) and antiviral drugs, neurotropic drugs, opiates, and NSAIDs before PRF were also recorded.
Data Processing
For ease of calculation, we conducted the following data preprocessing: In terms of sex, we marked “1” for female and “2” for male. The guide methods of the procedure include ultrasound recorded as “1,” CT + ultrasound recorded as “2,” radiography + ultrasound recorded as “3,” CT recorded as “4,” and radiography recorded as “5.” As for education level, we recorded illiteracy as “1,” elementary school level as “2,” junior high school as “3,” senior high school as “4,” and university and above as “5.” We divided different stages of HZ into three layers, which were represented by consecutive numbers: ≤ 1 month was recorded as “1,” between 1 and 3 months was recorded as “2,” and > 3 months was recorded as “3.” As for rash site, we recorded the cervical segment as “1,” head segments as “2,” lumbosacral segments as “3,” and thoracic segment as “4.” For the remaining features, we denoted “no” as “0” and “yes” as “1.” To describe effectiveness of the PRF treatment, we marked the negative as “0” and the positive as “1” according to general practices [7].
Statistical Analysis
The primary effectiveness endpoint was relief of > 50% in pain intensity as measured from baseline to 3 months after PRF, using an 11-point NRS. The outcome was effectiveness of the PRF treatment, which was defined as either “positive” or “negative.”
To compare the difference between the two outcomes to select significant features, we used the t-test for continuous variables and chi-square tests for categorical variables. Correlation analysis was applied to select strongly correlated variables. The least absolute shrinkage and selection operator (LASSO) method is suitable for reducing high-dimensional data and is applied to select the most useful prediction candidates, so LASSO regression was used to construct the model with the previously selected variables. We used tenfold cross-validation to obtain the optimal regularization parameter, i.e., lambda, by minimizing the sum of the least square plus shrinkage penalty. LASSO regression was performed using the glmnet package in R version 4.1.1. The performance of the prediction model was assessed by the area under the receiver operating characteristic (ROC) curve (AUC).