Microdiscectomy Is More Cost-effective Than a 6-Month Nonsurgical Care Regimen for Chronic Radiculopathy

Abstract

Background

A recent randomized controlled trial (RCT), performed by the authors, comparing early surgical microdiscectomy with 6 months of nonoperative care for chronic lumbar radiculopathy showed that early surgery resulted in improved outcomes. However, estimates of the incremental cost-utility ratio (ICUR), which is often expressed as the cost of gaining one quality-adjusted life year (QALY), of microdiscectomy versus nonsurgical management have varied. Radiculopathy lasting more than 4 months is less likely to improve without surgical intervention and may have a more favorable ICUR than previously reported for acute radiculopathy.

Question/purpose

In the setting of chronic radiculopathy caused by lumbar disc herniation, defined as symptoms and/or signs of 4 to 12 months duration, is surgical management more cost-effective than 6 months of nonoperative care from the third-party payer perspective based on a willingness to pay of less than CAD 50,000/QALY?

Methods

A decision analysis model served as the vehicle for the cost-utility analysis. A decision tree was parameterized using data from our single-center RCT that was augmented with institutional microcost data from the Ontario Case Costing Initiative. Bottom-up case costing methodology generates more accurate cost estimates, although institutional costs are known to vary. There were no major surgical cost drivers such as implants or bone graft substitutes, and therefore, the jurisdictional variance would be minimal for tertiary care centers. QALYs derived from the EuroQoL-5D were the health outcome and were derived exclusively from the RCT data, given the paucity of studies evaluating the surgical treatment of lumbar radiculopathy lasting 4 to 12 months. Cost-effectiveness was assessed using the ICUR and a threshold of willingness to pay CAD 50,000 (USD 41,220) per QALY in the base case. Sensitivity analyses were performed to account for the uncertainties within the estimate of cost utility, using both a probabilistic sensitivity analysis and two one-way sensitivity analyses with varying crossover rates after the 6-month nonsurgical treatment had concluded.

Results

Early surgical treatment of patients with chronic lumbar radiculopathy (defined as symptoms of 4 to 12 months duration) was cost-effective, in that the cost of one QALY was lower than the CAD 50,000 threshold (note: the purchasing power parity conversion factor between the Canadian dollar (CAD) and the US dollar (USD) for 2019 was 1 USD = 1.213 CAD; therefore, our threshold was USD 41,220). Patients in the early surgical treatment group had higher expected costs (CAD 4118 [95% CI 3429 to 4867]) than those with nonsurgical treatment (CAD 2377 [95% CI 1622 to 3518]), but they had better expected health outcomes (1.48 QALYs [95% CI 1.39 to 1.57] versus 1.30 [95% CI 1.22 to 1.37]). The ICUR was CAD 5822 per QALY gained (95% CI 3029 to 30,461). The 2-year probabilistic sensitivity analysis demonstrated that the likelihood that early surgical treatment was cost-effective was 0.99 at the willingness-to-pay threshold, as did the one-way sensitivity analyses.

Conclusion

Early surgery is cost-effective compared with nonoperative care in patients who have had chronic sciatica for 4 to 12 months. Decision-makers should ensure adequate funding to allow timely access to surgical care given that it is highly likely that early surgical intervention is potentially cost-effective in single-payer systems. Future work should focus on both the clinical effectiveness of the treatment of chronic radiculopathy and the costs of these treatments from a societal perspective to account for occupational absences and lost patient productivity. Parallel cost-utility analyses are critical so that appropriate decisions about resource allocation can be made.

Level of Evidence

Level III, economic and decision analysis.

Introduction

Lumbar radiculopathy, or neuralgia of the sciatic nerve, is characterized by paroxysmal attacks of pain in the buttock, back of the thigh, or in the leg or foot, and is often secondary to compression from disc herniation either at the L4/5 or L5/S1 levels. The treatments most commonly include physiotherapy, chiropractic manipulation, oral medications, image-guided steroid injections, or surgery, which all may incur considerable expense for health care systems and patients.

Cost-utility and cost-effectiveness analyses seek to quantify the costs incurred for various treatments incorporating the anticipated benefits of those treatments. The surgical treatment of lumbar radiculopathy has produced various incremental cost-utility ratio (ICUR) estimates, with some being below and some above the willingness-to-pay threshold (WTP) for healthcare systems worldwide [8, 11, 14, 15, 18, 20, 22]. Many of these studies evaluated acute radiculopathy, which has a favorable natural history and is distinctly different than radiculopathy lasting more than 4 months. Previous attempts to compare operative to nonoperative treatment are also difficult to interpret because patients who undergo surgery often differ from those who do not. The Spine Patient Outcomes Related Trial (SPORT), for example, demonstrated variability in the nonoperative treatment offered, as well as crossover and nonadherence. At 1 year, 43% of patients randomized to nonoperative care and only 59% of those randomized to surgery underwent surgical treatment; thus, a comparison based on intention-to-treat analysis found no difference between groups. However, an as-treated analysis found improvements in the surgically treated patients [13,18].

More recently, we compared surgical discectomy outcomes for chronic sciatica to nonsurgical strategies and found lasting patient-reported improvement at 1 year [2]. However, it is not clear whether operative or nonoperative interventions are more cost-effective. To date, we know of no prior studies that have compared the value or cost-utility of surgical treatment for chronic sciatica versus ongoing nonsurgical care. In the setting of a single payer, it is critical that cost-utility estimates are available to compare various treatments and interventions.

We therefore asked: In the setting of chronic radiculopathy caused by lumbar disc herniation, defined as symptoms and/or signs of 4 to 12 months duration, is surgical management more cost effective than nonoperative care?

Materials and Methods

Study Design and Setting

This economic evaluation is a cost-utility analysis of a randomized controlled trial (RCT) comparing nonsurgical and surgical treatment for chronic radiculopathy [2, 7]. The parent trial has been previously reported [2]. Briefly, consecutive patients referred to our clinic at a tertiary spine center were screened from February 2010 through August 2016. Patients were included if they were 18 to 60 years of age and had a posterolateral disc herniation at L4/5 or L5/S1 confirmed on MRI with unilateral radiculopathy lasting 4 to 12 months. Patients who had radiculopathy secondary to a herniation within the foramen; those who had far lateral disc, spinal stenosis, or deformity; patients who had previous lumbar surgery at the involved level; those who had received prior injections; or patients who had evidence-based physiotherapy were excluded from the trial. Patients were randomized by a computer-generated random number list in a 1:1 allocation to early surgery or nonsurgical care followed by surgery if needed. Randomization was stratified according to workers compensation status. Patients randomized to surgery underwent a single-level microdiscectomy procedure (open or minimal access approach) within 3 weeks of randomization either as a day surgery procedure or as a one-night stay at either the L4/5 or L5/S1 level. No patients underwent fusion procedures. The patients randomized to nonsurgical management were evaluated by a physiatrist/trial physician while on the surgical waiting list of the referred surgeon, which was a minimum 6-month wait. These patients received an individualized/tailored approach to education, oral medication, and exercise-based physiotherapy, and up to three epidural glucocorticoid injection(s)/nerve root block (Depomedrol, Pfizer) and 0.35% lidocaine. Patients could crossover to surgery if symptoms persisted after the standard wait of at least 6 months.

In the parent trial, 790 patients were screened for eligibility by telephone interview. Of these patients, 376 were eligible for a screening visit by a trial physician, and of these, 168 were eligible for enrollment. Forty patients declined. Therefore, 128 were included and underwent randomization, with 64 patients per group [2]. There were no differences in baseline characteristics between the surgical and nonsurgical groups. Regarding age, patients were 38 ± 8 years in the operative group versus 37 ± 12 years in the nonoperative group. In the operative group, 42% (27 of 64) of patients were females versus 39% (25 of 64) of patients in the nonoperative group. Duration of symptoms was 7.7 ± 2.9 months in the nonoperative group versus 7.3 ± 2.5 months in the operative group [2].

We created an analytic decision model using data from our RCT [1] to parameterize the model, supplemented with cost data from the Ontario Case Costing Initiative (OCCI), which is a bottom-up, microcase costing methodology providing the most accurate and robust patient-level data.

We used this model for two reasons: First, this approach is consistent with Sculpher’s recommendations about using RCT data for economic evaluations [6]. Briefly, there are substantial cost variations, and the external validity of single-center trials is limited. A model that accounts for variability in patient outcomes and varying degrees of potential costs acknowledges and quantifies the spectrum of potential values of an intervention. Second, using a decision tree accounts for potential crossover between treatment groups (either before or after the 6-month nonsurgical treatment phase). Probabilistic sensitivity, one-way sensitivity, and value of information analyses were performed.

Best- and worst-case scenarios using case-costing and patient-level data were considered. Patients in each group experienced a variety of potential adverse events within the 2-year time, including dural tears (requiring hospital admission rather than standard same-day discharge), superficial wound infection (as per Centers for Disease Control and Prevention definition), postoperative adjacent-level condition, and new onset postoperative neuropathic pain. Specifically, reherniation requiring revision surgery occurred in two patients from the operative group and one patient from the nonoperative group. These instances represented the worst-case scenario where more expensive surgery was undertaken, increasing overall costs with a poorer outcome overall for the patients. Typically, a systematic review will inform event probabilities with far greater numbers of patients for comparison and evaluation. Chronic radiculopathy has received little attention, and previous trials have not specifically defined chronic radiculopathy lasting 4 to 12 months [8, 21, 22]. Our data were rigorously collected at appropriate timepoints, and patients within the trial reported a wide variety of outcome scores. As illustrated in our original publication, chronic radiculopathy represents a distinct group of patients who are unlikely to experience meaningful improvements without surgical management after 4 months [2]. Although there are limitations in considering a single-site, small but adequately powered RCT, we believe that the broad variety of patient outcomes and inputs on costs within this study adequately account for the breadth in which the various clinical possibilities could occur.

Study Perspective, Time Horizon, and Discount Rate

The patients were followed for 2 years, but all costs were incurred within 1 year; consequently, no discounting of costs was necessary, following the method of Glick et al. [9]. However, the second-year quality-adjusted life years (QALYs) were discounted following the National Agency for Drugs and Technologies in Health guidelines using a 1.5% per annum discount rate [4]. It is conceivable to ask whether the 2-year period is adequate. According to Drummond [7], the required time horizon for an economic evaluation is the period during which costs and health outcomes differ between the treatment groups. Further, Drummond [7] argued that if there are mortality differences associated with the treatment received, the time horizon should be a patient's expected lifetime. In our study, based on expert opinion, the 2-year period was adequate because there were no mortality differences between the two treatment arms.

Choice of Health Outcome

The health outcome was QALYs via EuroQoL-5D (EQ-5D) scores. In the parent RCT, the primary outcome was intensity of leg pain at 6 months after enrollment on a VAS (0 = no pain; 10 = worst pain); however, unlike EQ-5D scores, VAS leg pain scores have not routinely been used to derive health states. The EQ-5D was collected at baseline, 6 weeks, and 3, 6, 12, and 24 months after enrollment. The EQ-5D asks five critical questions about patient functional living, including mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. We collected health-related quality of life data from participants’ EQ-5D scores at baseline and subsequent follow-up periods using generated health state estimates (Table 1). We used value sets for the United States to convert health states to health utility scores. The health utility scores were converted into QALYs following Glick et al. [9]. First, the mean utility score was calculated by adding the score at the beginning of each period to the end and dividing it by two. This figure was then multiplied by either 0.25 (baseline to 3 months and 3 to 6 months) or 0.5 (6 to 12 months). The sum of QALYs for baseline to 3 months, 3 to 6 months, and 6 to 12 months constitutes the QALYs for the first year. Second-year QALYs (12 to 24 months) were discounted using a rate of 1.5% [4]. We used the last observation carried forward imputation method to account for missing health utility scores. The sum of QALYs for years 1 and 2 constitute the overall QALYs.

Table 1.

Comparison of EQ-5D values between operative and nonoperative groups by time

Variable and timepoint of assessment	Nonoperative		Operative		Treatment effect (95% CI)	p value
	Number of patients	Mean ± SE	Number of patients	Mean ± SE
EQ-5D
Baselinea	64	0.48 ± 0.20	63	0.48 ± 0.22	-0.004 (0.077-0.070)	0.92
6 weeks	57	0.61 ± 0.02	59	0.75 ± 0.03	0.136 (0.067- 0.204)	0.001
3 months	50	0.63 ± 0.03	51	0.76 ± 0.03	0.130 (0.057- 0.202)	0.001
6 months	54	0.63 ± 0.02	51	0.77 ± 0.03	0.146 (0.074- 0.217)	0.001
12 months	47	0.68 ± 0.03	51	0.78 ± 0.03	0.105 (0.032- 0.178)	0.005
24 months	41	0.76 ± 0.03	48	0.78 ± 0.03	0.019 (-0.057- 0.094)	0.63

Means are derived from mixed-model repeated-measures analyses using a random effect for participant and a compound symmetry assumption on the correlation of residuals; the dependent variable was the score at each timepoint; fixed effects included the baseline score, treatment, and time; time was a categorical variable.

^aNote that baseline data were included in the model as a covariate but not derived from the model.

Health Resource Use and Costs

Costs included medications for 6 months, rehabilitation, physiotherapy, injections, bundled cost of surgery, physician fee, and cost of hospital stay in patients with complications from surgery. All costs incurred within the trial for patient treatment were obtained from the Ontario Health Insurance Plan. Single-payer, central governmental insurance is the same throughout Canada and comparable to other tax-based systems for healthcare payment. The nonoperative care group’s relevant costs were associated with drugs, injections, rehabilitation, and physiotherapy. Importantly, patients were not eligible for our randomized trial if they had received prior nonsurgical management for their condition; however, patients from both groups were taking medication at baseline (Table 2) [2]. Each patient within the nonoperative group was assessed independently by a rehabilitation expert (KS, TM), and their nonoperative treatment was tailored based on symptom severity. We calculated three scenarios for medication use: best case, medium case, and worst case. A similar approach was used for rehabilitation costs. Full details of physiotherapy usage and injection data were captured (Fig. 1). The costs associated with patients treated with early surgery at the time of randomization included the physician fee for surgery; the bundled cost of surgery, which was the mean of the least and most expensive encounters and the associated uncertainty; and the cost of a standard hospital stay [23]. The bundled cost-of-surgery data came from the case-costing center of the Health Science Centre (the OCCI). It consisted of labor and supply costs for food services, surgical inpatient services, operation and operating room supplies, postanesthesia care, clinical laboratory costs, medical imaging, labor and supplies for pharmacy, postoperative medications, and physiotherapy, and it includes an estimation for overhead expenses, administrative, and infrastructure costs. In the event of a crossover to surgery, the costs associated with the surgery were applied. All costs were measured in CAD and were in 2019 prices. The purchasing power parity conversion factor between the Canadian dollar (CAD) and the US dollar (USD) for 2019 was 1 USD = 1.213 CAD. There was no discounting of costs because all costs were incurred within 1 year (Table 3) [4].

Table 2.

Medication use at baseline^a

Medication	Nonoperative (n = 64)	Operative (n = 64)
NSAID	55 (35)	42 (27)
Cyclooxygenase-2 inhibitor	6 (4)	13 (8)
Narcotics	48 (31)	53 (34)
Muscle relaxant	6 (4)	0 (0)
Neuromodulator (gabapentin or pregabalin)	28 (18)	25 (16)
Antidepressant	8 (5)	28 (18)
Other	13 (8)	20 (13)
None	11 (7)	13 (8)

Data presented as % (n).

^aNote that patients could be taking more than one medication.

Fig. 1.

Nonoperative patient physiotherapy usage and aggregate injection administration with responses to treatment.

Table 3.

Model parameters

Parameter	Treatment group at randomization
	Surgery, mean ± SE	Nonoperative care, mean ± SE
QALYsa
Baseline	0.49 ± 0.03	0.49 ± 0.03
Year 1 (0-12 months)	0.73 ± 0.02	0.61 ± 0.02
0-3 months	0.16 ± 0.008	0.14 ± 0.006
3-6 months	0.19 ± 0.006	0.15 ± 0.006
6-12 months	0.38 ± 0.01	0.32 ± 0.01
Year 2 (12-24 months)	0.76 ± 0.03	0.69 ± 0.03
Discounted (12-24 months)	0.74 ± 0.02	0.68 ± 0.02
Total QALYs (years 1 and 2)	1.48 ± 0.04	1.30 ± 0.05
Costs in CAD
Medication costs for 6 months		281 ± 196
Rehabilitation		538 ± 242
Physiotherapy		324 ± 117
Epidural steroid injection		400 ± 115
Bundled cost of surgeryb	2890 ± 362
Physician fee for surgery	1007
Hospital stay	5460 ± 1092
Probabilitiesc
Crossover ≤ 6 months	0.053 ± 0.01	0.016 ± 0.003
Crossover > 6 months		0.291 ± 0.06
Complication after surgery	0.075 ± 0.02	0.239 ± 0.05

^aOur computation was based on data from the randomized controlled trial.

^bThe bundled cost of surgery consisted of the costs of labor and supply for food services, surgical inpatient services, operation and operating room supplies, postanesthesia care, clinical laboratory, medical imaging, labor and supplies for pharmacy, postoperative medications, and physiotherapy; all costs were measured in CAD and were in 2019 prices. The purchasing power parity conversion factor between CAD and the US dollar for 2019 was 1 USD = 1.213 CAD.

^cProbabilities were derived from data from the randomized controlled trial; the 0.053 probability of crossover ≤ 6 months could be interpreted as a 5.3% likelihood of a crossover.

Probabilities

All probabilities used to parameterize the analytic decision model were estimated using patient-level data from the RCT [2]. These included the probability of crossover before and after 6 months, which differed between the two treatment arms, and the probability of surgery-related complications (adverse events), which again differed between the treatment arms. Based on the patient-level data, 8% (DF of 64) of patients in the early surgery group had an adverse event. In comparison, in the nonoperative care group, 9% (6 of 64) of patients had adverse outcomes. Additionally, 3% (2 of 64) of patients in the nonoperative care group crossed over to early surgery within 6 months, and 11% (7 of 64) of patients in the early surgery group crossed over to nonoperative care. After the same period, 34% (22 of 64) of patients in the nonoperative group crossed over to surgery. The trial was performed at a single institution and was originally powered to detect a difference in leg pain scores. EQ-5D scores were used to estimate QALYs. Cost-utility estimates were calculated both at 1 and 2 years to illustrate the uncertainty regarding patient outcomes at the 1-year mark.

Assessing Cost-effectiveness

A decision tree served as the vehicle for the cost-utility analysis (Fig. 2). It illustrated the possible combinations of treatments, crossovers, complications, costs, and health outcomes associated with each treatment. For each arm of the tree, the health resources used were identified, measured, and valued. Cost-effectiveness was assessed using the ICUR, defined as $\text{ICUR}=\frac{\Delta C}{\Delta E}$ , where $\Delta E$ is the incremental QALYs and $\Delta C$ is the incremental costs. WTP thresholds vary dramatically across countries and even some states/provinces in a global sense; however, less than 50,000/QALY is generally accepted for Organisation for Economic Co-operation and Development (OECD) countries. Therefore, in keeping with usual practice, we similarly defined cost-effectiveness as an intervention costing less than CAD 50,000/QALY (USD 41,220).

Fig. 2.

This decision tree illustrates the various potential clinical pathways in each treatment group.

Sensitivity Analyses

We conducted two sensitivity analyses: a probabilistic sensitivity analysis and two one-way sensitivity analyses. In the probabilistic sensitivity analysis, we allowed key parameters to vary simultaneously, and the results were evaluated [7]. The probabilistic sensitivity analysis provides an opportunity to measure the impact of uncertainties about the parameters used in the model on the model outcomes. Instead of changing the value of one parameter and analyzing its effects, the probabilistic sensitivity analysis allows all relevant model parameters to be varied simultaneously by fitting appropriate probability distributions to these parameters. All potential healthcare outcomes from patient-level data and a spectrum of potential cost outcomes populates this analysis.

The cost uncertainties were modeled using a gamma distribution, QALYs were modeled using a lognormal distribution, and probabilities were modeled using a beta distribution [3]. The hyperparameters required for these distributions were estimated using the method of moments, using the mean and standard errors of the parameters as inputs (Table 3). A Monte Carlo simulation was used to generate 10,000 simulated trials. The results were used to calculate the 95% confidence interval around the ICUR using the percentile method [10]. A cost-effectiveness acceptability curve, which shows how the probability that an intervention is cost-effective changes at various WTP levels, and cost-effectiveness planes were used to summarize the simulations’ results.

One feature that separates our study from others is our treatment of the crossover probabilities. In other words: Are our results sensitive to changes in the crossover probabilities? The one-way sensitivity analysis provided us with an opportunity to check this. It also served as a robustness assessment of our results. In the one-way sensitivity analyses, the probability of crossover at 6 months and after 6 months was allowed to vary from 0.01 to 0.90. Given the major confounding factor for this trial and other trials involving sciatica are crossover from one treatment arm to the other, we specifically chose this parameter. Further, the analysis, including the probabilistic sensitivity analysis, was repeated using health outcomes associated with only 1 year of follow-up.

Ethical Approval

We obtained ethical approval for this study from the institutional review board at University of Western Ontario, London, Canada (review number 16000).

Results

Summary of Results

This study found that early surgical treatment was more cost-effective than nonoperative care. Early surgery was associated with better expected QALYs than nonoperative care, but it had higher expected costs. The incremental costs were sensitive to changes in the probability of crossover from nonoperative care to surgical treatment, but the conclusions about cost-effectiveness did not change.

Cost-effectiveness Results

Early surgical treatment of patients with chronic lumbar radiculopathy (defined as symptoms of 4 to 12 months duration) was cost-effective in that the cost of one QALY was lower than CAD 50,000 (USD 41,220). We estimated that early surgery resulted in costs of CAD 4118 (95% CI 3429 to 4867), compared with CAD 2377 (95% CI 1622 to 3518) for nonoperative management. However, the expected QALYs were higher in the surgical group than in the nonoperative group for the subsequent follow-up periods. This is to say that the health states or status of patients treated with surgery were better than those treated nonsurgically. These QALYs were the outcomes used in the decision tree. There were QALY gains until 2 years (using EQ-5D patient-reported outcomes), but costs were stable after 1 year (Fig. 3).

Fig. 3.

Graph demonstrating differences in EQ-5D scores at various points over 2 years, which illustrates the method for estimating the difference in QALYs gained. The difference is outlined by the dashed green lines; costs remained relatively constant after 1 year, reducing the cost-utility ratio over time. A color image accompanies the online version of this article.

The early surgical treatment arm had 1.48 expected QALYs (95% CI 1.39 to 1.57), and the nonoperative treatment arm had 1.30 expected QALYs (95% CI 1.22 to 1.37) (Table 4). The incremental QALY was 0.18 (95% CI 0.06 to 0.30). Because the 95% CI of the incremental QALYs excluded 0, the difference in QALYs between the two groups was statistically significant at the 5% significance level. The expected cost in the early surgical group was CAD 4161 (95% CI 3429 to 4867) and CAD 3113 (95% CI 1622 to 3518) in the nonoperative group, with an incremental cost of CAD 1048 (95% CI 590 to 2678). Early surgical care appeared to be costly but resulted in better health outcomes. The ICUR was CAD 5822 (USD 4800) per QALY gained (95% CI 3029 to 30,461).

Table 4.

Cost-effectiveness results at 2 years comparing operative and nonoperative care

Parameter	Reference		PSA (n = 10,000 Monte Carlo simulations)
	Operative	Nonoperative	Operative (95% CI)	Nonoperative (95% C)
Expected costs in CAD	4161	3113	4118 (3429-4867)	2377 (1622-3518)
Expected QALYs	1.48	1.30	1.48 (1.39-1.57)	1.30 (1.22-1.37)
Incremental QALYs	0.18		0.18 (0.06-0.30)
Incremental cost in CAD	1048		1742 (590-2678)
ICUR in CAD	5822 per QALY gained		9751 per QALY gained (3029-30,461)
The probability that surgery is cost-effective at a WTP of CAD 50,000 per QALY			0.99

All costs were measured in Canadian dollars and were in 2019 prices; the purchasing power parity conversion factor between CAD and the US dollar for 2019 was 1 USD = 1.213 CAD; PSA = probabilistic sensitivity analysis; WTP = willingness-to-pay threshold.

Methodologic Uncertainty

The threshold for WTP was allowed to vary from CAD 0 to CAD 50,000. The crossover point where surgical care was more cost-effective than nonsurgical care occurred at about CAD 10,000 (Fig. 4). The results of 10,000 Monte Carlo simulations demonstrate 9848 of 10,000 points were below the WTP threshold (Supplementary Fig. 1; http://links.lww.com/CORR/A633).

Fig. 4.

This cost-effectiveness acceptability curve shows how the probability that surgery is cost-effective changes with different values of the willingness-to-pay threshold. The purchasing power parity conversion factor between the Canadian dollar and the US dollar for 2019 was USD 1 = CAD 1.213. A color image accompanies the online version of this article.

Also, the results from the probabilistic sensitivity analysis, based on the 1-year data, mirrored results already reported, particularly the probability that surgical treatment was cost-effective at the stated WTP of 0.93 (9339 of 10,000 simulations) (Table 5). (Note the wider confidence interval compared with Table 4 reflects the greater uncertainty in health states at 1 year compared with the cumulative improvement over 2 years.)

Table 5.

Cost-effectiveness results for 1 year of follow-up

Parameter	Reference		PSA (n = 10,000 Monte Carlo simulations)
	Operative	Nonoperative	Operative (95% CI)	Nonoperative (95% CI)
Expected costs in CAD	4161	3113	4118 (3429-4867)	2377 (1622-3518)
Expected QALYs	0.73	0.61	0.73 (0.69-0.78)	0.61 (0.52-0.72)
Incremental QALYs	0.12		0.12 (0.01-0.22)
Incremental cost in CAD	1048		1742 (590-2678)
ICUR in CAD	8792 per QALY gained		14,517 per QALY gained (1431-69,872)
The probability that surgery is cost-effective at a WTP of CAD 50,000 per QALY			0.93

All costs were measured in Canadian dollars (CAD) and were in 2019 prices; the purchasing power parity conversion factor between CAD and the US dollar (USD) for 2019 was 1 USD = 1.213 CAD; PSA = probabilistic sensitivity analysis; WTP = willingness-to-pay threshold.

One-way Sensitivity Analyses

The one-way sensitivity analyses confirmed the main finding that surgical treatment was more cost-effective than nonoperative management. However, if the probability of crossover from nonoperative care to early surgery before 6 months exceeded 0.34 (approximately one-third of patients), the expected costs associated with surgical treatment became smaller than the one associated with nonoperative care (Fig. 5).

Fig. 5.

One-way sensitivity analysis for different values of the probability of crossover at 6 months. The purchasing power parity conversion factor between the Canadian dollar and the US dollar for 2019 was USD 1 = CAD 1.213. A color image accompanies the online version of this article.

Similarly, after 6 months, a probability of crossover greater than approximately 0.50 (or one-half of patients) was associated with increased costs for patients randomized to nonoperative care (Fig. 6).

Fig. 6.

One-way sensitivity analysis for different values of the probability of crossover after 6 months. The purchasing power parity conversion factor between the Canadian dollar and the US dollar for 2019 was USD 1 = CAD 1.213. A color image accompanies the online version of this article.

Discussion

Chronic lumbar radiculopathy is a common clinical presentation, although it has not been thoroughly evaluated with a consistent definition with respect to the quantity of time a patient experiences leg pain. More patients are presenting with this condition, and even though early surgical intervention has been shown to be more clinically effective, it is critical to quantify the cost utility of early surgery. The current analysis demonstrated that early surgery for chronic lumbar radiculopathy is cost-effective compared with nonoperative care. When comparing our study to the prior cost-effectiveness analyses performed on the topic of lumbar radiculopathy, the ICUR of CAD 5822 per QALY is well below prior estimates published for patients with acute disc herniation [15, 19, 21]. After accounting for uncertainty in the sensitivity analysis, the result remains cost-effective, well below the WTP in OECD countries. [13]

Limitations

The main limitation of this work is that the cost-utility analysis was performed from a small, single-centered RCT. Although the cost and QALY estimates are felt to be robust and representative of our sample, we acknowledge the external validity problems of this trial. Although adverse events occurred in both groups at a moderate percentage, there exists the possibility that major adverse events have been missed. Generally, a systematic review of the evidence will identify these severe adverse events and may alter the ICUR, driving costs up while leading to poorer patient outcomes. Future work should focus on generating more effectiveness research on chronic sciatica along with microcase costing data so healthcare payers can be more informed about the impact of resource allocation decisions.

Nonsurgical care was also not standardized among all patients. Clinically, this is challenging to accomplish given a clinical variance in symptom severity from different patients, although, the cost estimations based on worst-/best-case scenarios may have been more or less, ultimately causing an unpredictable effect on the ICUR estimate. Costs/QALY estimates are only as strong as the raw data that populate the model fields.

An additional limitation is that the analysis did not include a societal estimate for cost-effectiveness/utility. This, in addition to the single-center nature of the trial, limits the external validity of our results. The productivity losses of patients were not assessed because hourly wages, costs of replacement workers, and total time absent from work were not captured. When making cost-utility estimates with results alongside a clinical trial, this information is critical. The results omit the probable productivity gains by the early surgical group, while failing to account for the nonoperative care group’s ongoing productivity losses. The ICUR is likely lower because those undergoing surgery are more likely to return to work, which can be inferred based on the higher rate of employment in the original trial [2].

Cost-effectiveness of Early Surgery for Chronic Lumbar Radiculopathy

We found that surgical treatment remained more cost-effective than nonoperative care, even when patients initially treated without surgery crossed over and underwent surgery later in their clinical course. This information is important for both the treating surgeon and the healthcare payer. Clinically speaking, offering patients medications, injections, and ongoing nonsurgical management therapies, although less costly, is less likely to be effective. Often, one or possibly two nonsurgical treatments are trialed before consenting for surgery. Although both groups had originally trialed a variety of medications before randomization, no patient had any injections or active physiotherapy. Our results demonstrate that prescribing these nonsurgical options may not be best from a resource allocation perspective after pain has persisted for more than 4 months or up to and including 1 year. Persisting with nonsurgical care that is costly and less effective should be questioned by payers, and resources potentially allocated to ensure faster access to surgical management [16].

Prior trials randomizing patients to treatment for sciatica have had issues with crossover [11]. Cost-utility/effectiveness results have therefore suffered when considering as-treated results since patient-reported outcomes were similar. The favorable cost-utility estimate in this study is likely low for two reasons: first, very few patients crossed over at the 6-month mark, driving better clinical results of surgery, and second, because surgery occurred early, there were fewer pre- and postoperative costs for patients undergoing surgery. These findings are a unique aspect of our paper, and further study is required to more specifically evaluate the treatment of radiculopathy that has lasted 4 to 12 months, ideally with bottom-up microcase-costing to ensure robust cost-utility estimates. [5]

Prior estimates of the cost-utility/effectiveness of spine surgery provide some context for both the importance and impact of our study [17]. There is variability in cost-utility estimation based on the methodology, and often a 5-year or “lifetime” horizon is used with an assumption of sustained patient improvement beyond the study period [1, 12]. Our estimate for cost utility is based on a 2-year endpoint and makes no assumptions about clinical outcome/future costs beyond the reported timeframe. When it is assumed that patients continue to experience health benefits beyond the study period, the cost-utility ratio often decreases, which makes for a fragile assumption and potentially inaccurate conclusions about cost utility. We believe that by restricting our analysis to the 2-year endpoint, we have avoided some of the potential problems of prior cost-utility research in spine surgery.

Conclusion

Early surgery for chronic lumbar radiculopathy is cost-effective compared with nonsurgical management. Nonsurgical care is less effective and less costly, and the tradeoff for more costly but more effective surgical care is justified by our low estimate for cost utility. Future studies analyzing surgical treatment for chronic radiculopathy with accompanying robust cost estimates should incorporate societal impacts of care. Accounting for lost productivity may lead to lower cost-utility estimates and generate stronger justification for early surgery in patients with prolonged discogenic radiating leg pain.