Systematic Review and Neural Network Analysis to Define Predictive Variables in Implantable Motor Cortex Stimulation to Treat Chronic Intractable Pain

Implantable motor cortex stimulation (iMCS) has been performed for >25 years to treat various intractable pain syndromes. Its effectiveness is highly variable and, although various studies revealed predictive variables, none of these were found repeatedly. This study uses neural network analysis (NNA) to identify predictive factors of iMCS treatment for intractable pain. A systematic review provided a database of patient data on an individual level of patients who underwent iMCS to treat refractory pain between 1991 and 2017. Responders were defined as patients with a pain relief of >40% as measured by a numerical rating scale (NRS) score. NNA was carried out to predict the outcome of iMCS and to identify predictive factors that impacted the outcome of iMCS. The outcome prediction value of the NNA was expressed as the mean accuracy, sensitivity, and specificity. The NNA furthermore provided the mean weight of predictive variables, which shows the impact of the predictive variable on the prediction. The mean weight was converted into the mean relative influence (M), a value that varies between 0 and 100%. A total of 358 patients were included (202 males [56.4%]; mean age, 54.2 ±13.3 years), 201 of whom were responders to iMCS. NNA had a mean accuracy of 66.3% and a sensitivity and specificity of 69.8% and 69.4%, respectively. NNA further identified 6 predictive variables that had a relatively high M: 1) the sex of the patient (M = 19.7%); 2) the origin of the lesion (M = 15.1%); 3) the preoperative numerical rating scale score (M = 9.2%); 4) preoperative use of repetitive transcranial magnetic stimulation (M = 7.3%); 5) preoperative intake of opioids (M = 7.1%); and 6) the follow-up period (M = 13.1%). The results from the present study show that these 6 predictive variables influence the outcome of iMCS and that, based on these variables, a fair prediction model can be built to predict outcome after iMCS surgery. Perspective: The presented NNA analyzed the functioning of computational models and modeled nonlinear statistical data. Based on this NNA, 6 predictive variables were identified that are suggested to be of importance in the improvement of future iMCS to treat chronic pain.