A novel use of the Spine Tango registry to evaluate selection bias in patient recruitment into clinical studies: an analysis of patients participating in the Lumbar Spinal Stenosis Outcome Study (LSOS)

Abstract

Purpose

Patients enrolled in clinical studies typically represent a sub-set of all who are eligible, and selection bias may compromise the generalizability of the findings. Using Registry data, we evaluated whether surgical patients recruited by one of the referring centres into the Lumbar Spinal Stenosis Outcome Study (LSOS; a large-scale, multicentre prospective observational study to determine the probability of clinical benefit after surgery) differed in any significant way from those who were eligible but not enrolled.

Methods

Data were extracted for all patients with lumbar spinal stenosis registered in our in-house database (interfaced to Eurospine’s Spine Tango Registry) from 2011 to 2013. Patient records and imaging were evaluated in relation to the admission criteria for LSOS to identify those who would have been eligible for participation but were not enrolled (non-LSOS). The Tango surgery data and Core Outcome Measures Index (COMI) data at baseline and 3 and 12 months after surgery were analysed to evaluate the factors associated with LSOS enrolment or not.

Results

514 potentially eligible patients were identified, of which 94 (18%) were enrolled into LSOS (range 2–48% for the 6 spine surgeons involved in recruiting patients) and 420 (82%) were not; the vast majority of the latter were due to non-referral to the study by the surgeon, with only 5% actually refusing participation. There was no significant difference in gender, age, BMI, smoking status, or ASA score between the two groups (p ≥ 0.18). Baseline COMI was significantly (p = 0.002) worse in the non-LSOS group (7.4 ± 1.9) than the LSOS group (6.7 ± 1.9). There were no significant group differences in any Tango surgery parameters (additional spine patholothegies, operation time, blood loss, complications, etc.) although significantly more patients in the non-LSOS group had a fusion procedure (38 vs 18% in LSOS; p = 0.0004). Postoperatively, neither the COMI nor its subdomain scores differed significantly between the groups (p > 0.05). Multiple logistic regression revealed that worse baseline COMI (p = 0.021), surgeon (p = 0.003), and having fusion (p = 0.014) predicted non-enrolment in LSOS.

Conclusion

A high proportion of eligible patients were not enrolled in the study. Non-enrolment was explained in part by the specific surgeon, worse baseline COMI status, and having a fusion. The findings may reflect a tendency of the referring surgeon not to overburden more disabled patients and those undergoing more extensive surgery with the commitments of a study. Beyond these factors, non-enrolment appeared to be somewhat arbitrary, and was likely related to surgeon forgetfulness, time constraints, and administrative errors. Researchers should be aware of potential selection bias in their clinical studies, measure it (where possible) and discuss its implications for the interpretation of the study’s findings.