Study Design. Patients with low back pain were asked to recall the pain and impaired functioning that they reported 5-10 years previously as part of the National Low Back Pain prospective follow-up study. In 1998, patients completed an additional follow-up.
Objectives. To compare outcomes using patient-recalled data and prospectively collected data from patients with low back pain and to identify simple, symptom-specific questions that yield reliable responses over an extended period of time.
Summary of Background Data. Outcome assessment based on patient recall may be influenced by a patient's age, gender, reporting tendency, and current health status. The impact of data collected retrospectively on outcome analyses in spinal patients has not been addressed.
Methods. Patients enrolled in the National Low Back Pain study from 1986 to 1991 completed a self-administered questionnaire at their initial visit. A sample was interviewed by telephone in 1996 and asked to recall pain characteristics and impaired functioning reported at initial examination. A 10-year follow-up (1998) on current health status was conducted by mail. The 1998 follow-up response was separately compared with recalled and initial responses, such that two patient outcome status values were calculated for each question. Agreement was evaluated using Cohen's kappa.
Results. The follow-up evaluation was completed by 144 patients, with a mean interval of 9.4 years. The overall simple kappa was 0.37, indicating fair agreement between outcomes based on initial and recalled accounts of pain. Questions on location of pain had kappa values of 0.12-0.58, radicular symptoms 0.28-0.48, and severity of pain 0.11-0.30.
Conclusions. Fair to moderate agreement was found between outcomes determined by recalled versus initial reports. Accuracy was greatest for queries on frequency, location of pain, and activities affecting pain. Discrepancies were noted for queries on severity of pain, with error bias toward less pain when using the recalled data. Careful selection of questions may yield more accurate outcome measures.