Cherkin 2018.

Study characteristics
Methods	Cluster‐randomised trial, USA
Participants	2138 patients visited the intervention clinics and 2571 the control clinics. Six primary care clinics were pair randomised, three to training in the STarT Back strategy and three to serve as controls.
Interventions	The STarT Back risk‐stratification strategy matches treatments for LBP to physical and psychosocial obstacles to recovery using patient‐reported data (the STarT Back Tool) to categorize patients’ risk of persistent disabling pain.   Intervention features Multiple simple feedback (one PROM at multiple times) PROM(s) used as intervention: STarT Back tool (back pain) Constructs measured: Symptoms, Functioning Instrument categories/domains: Domain/Disease specific (back pain)   Administration features Where PROMs administered: Non‐clinical setting How administered: Interviewer‐administered Format of PROMs questionnaire(s): Electronic   Feedback features Format of PROMs feedback: Electronic How often information fed back: 3 times (baseline, 2 months, and 6 months) Who information fed back to: Clinicians Information fed back: Scores, Interpretation guidance, Management recommendations
Outcomes	Main outcomes: back‐related physical function and pain severity Other outcomes: healthcare utilisation.
Notes	Funding for this trial was provided by the Patient Centered Care Research Institute (“Evaluation of a Patient‐Centered Risk Stratification Method for Improving Primary Care for Back Pain”: Contract #398) and by the National Center for Complementary and Integrative Health/NIH (“Implementing Evidence‐Based Treatments for Persistent Back Pain into Primary Care”: Grant No. R21AT0007326). Martin Levine, Diane Piekara, and Pam Rock received support to participate in the quality improvement activities from Group Health. Nadine E Foster, an NIHR Senior Investigator, and Jonathan C. Hill were supported through an NIHR Research Professorship (NIHR‐RP‐ 011‐015) awarded to Nadine Foster. The study recruited between March 2013 and December 2015. The authors do not report conflicts of interest information.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Trial biostatistician randomly assigned participants using computer‐generated system.
Allocation concealment (selection bias)	High risk	Due to cluster‐randomised design not possible to conceal allocation from clinicians.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Due to nature of intervention not possible to blind patients and personnel.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Due to nature of the intervention blinding of outcomes not possible: PROM used for feedback also used to assess outcome, patients were aware they received the intervention.
Baseline outcome measurements similar	Low risk	Similar results of baseline outcomes in Table 1.
Baseline characteristics similar	Low risk	Characteristics of both groups similar.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Interviewers used to collect outcome data via telephone to reduce missing data.
Was study protected against contamination	Low risk	Intervention was delivered in separate clinics.
Selective reporting (reporting bias)	Low risk	All outcome measurements mentioned in methods section was reported in results.