Introduction: Low back pain (LBP) is the most common form of musculoskeletal pain, accounting for more years lived with disability than any other condition. In primary care, the majority of cases are labelled ‘nonspecific’ or ‘mechanical’, umbrella terms belying several etiologies with undifferentiated management approaches and outcomes. Current recommendations call for stratified care.
We investigated whether distinct, clinically-relevant classes (subgroups) of patients within a primary care interprofessional LBP program could be identified, and whether class could predict post-care pain and disability outcomes.
Methods: Prospective patients completed intake questionnaires including sociodemographic, health, and psychosocial characteristics. Low back pain pattern was classified by clinicians using the CORE Back Tool.
The primary outcome measures were numeric pain rating scale (NPRS) and Oswestry Disability Index (ODI), completed at intake and six-month follow-up. Binary outcomes were defined at follow-up for clinically important improvement (CII; NPRS: 2pt, ODI: 10pt) and minimal pain/disability (NPRS: 0-3, ODI: 0-20).
Analysis included patients with complete NPRS and ODI data; missingness in other measures was addressed with multiple imputation. Respondents were randomly allocated to equal-sized discovery and prediction samples.
Discovery: A latent class model was developed considering intake factors and six-month outcomes. The optimal number of classes was established based on clinical reasoning and several fit statistics.
Prediction: Most likely class membership was determined for each patient based on findings from the above latent class model. Outcome prediction models were then generated for clinically-important improvement and minimal pain/disability. Performance was assessed using c-statistic.
Results: 1330 participants (58% female, mean age 53) were allocated to discovery (N=667) and prediction (N=663) samples.
Discovery: Four classes emerged: 1) “sciatica” (16% of the sample; younger, leg-dominant pain), 2) “osteoarthritis” (18%; older, extension-aggravated back pain or intermittent leg pain), 3) mechanical “discogenic pain” (34%; younger, back-dominant pain), and 4) “persistent symptoms” (32%; severe symptoms and poor mental/physical health).
In the overall sample, CII rates were 57% for pain and 47% for disability, and 38% and 46% achieved minimal pain and disability at six month follow-up. When examined by class, outcomes varied considerably. “Sciatica”: substantial improvement (NPRS: 100% CII, 87% minimal pain; ODI: 94% CII, 84% minimal disability). “Osteoarthritis”: moderate improvement (NPRS: 58% CII, 29% minimal; ODI: 39% CII, 28% minimal). “Discogenic”: low levels of pain and disability at baseline and follow-up (NPRS: 59% CII, 53% minimal; ODI: 40% CII, 70% minimal disability). “Persistent symptoms”: poor outcomes (NPRS: 33% CII, 2% minimal; ODI: 29% CII, 1% minimal).
Prediction: The class-based prediction model had relatively good performance. Overall c-statistics were 0.72/0.67 for six-month CII in NPRS/ODI, and 0.65/0.78 for minimal pain/disability at six months.
Discussion: This study shows that based on a few select features, LBP patients can be pragmatically grouped into four identifiable clinical phenotypes that have distinct trajectories of pain and function. This suggests that LBP patients should not be viewed as a homogeneous group. Class-dependent outcome prediction achieved good performance. Simple identification of these four LBP classes by frontline clinicians may enable combined patient-clinician informed discussions about LBP prognosis in the absence of more complex point-of-care prediction tools.