We employed a retrospective cohort design to analyze health insurance claims data provided by the health network. The datasets consisted of paid claims containing information on providers, health plan members, and clinical encounters. The data were analyzed primarily for descriptive statistics, trends in expenditures, and exposure factors affecting expenditures. The study was conducted in accordance with all ethical research standards and approved by the Institutional Review Board, Southern California University of Health Sciences.
Population
The study population included adult patients aged 18 years of age or older with claims for chiropractic care delivered through the ChiroCare Network, with health plan coverage for chiropractic services. We included health plan members with a new primary diagnosis of LBP (as defined by any one of a selected constellation of diagnosis codes) and an acute or sub-acute episode duration. The rationale for restricting the study to acute and sub-acute care was to differentiate newly diagnosed patients from those with chronic pain, for whom the approach to care may differ.
Inclusion and exclusion criteria
The time period covered by the cost analysis ranged between 07/01/2016 and 12/22/2017. New patients 18 years of age or older with at least two claims for a primary diagnosis of LBP within 90 days and a treatment start date occurring 07/01/2016 and later were included in the analyses. We utilized a look-back period of 180 days to exclude patients with a history of low back pain prior to 07/01/2016. Patients with comorbid diagnosis of cancer, trauma, drug abuse, or infections of the spine were also excluded. Patients with chronic low back pain (i.e., with episode duration greater than 90 days) were also excluded. Prior to analysis, the dataset was subjected to extensive data cleaning operations intended to eliminate erroneous data (e.g., duplicate claims) and enhance validity, application of exclusion criteria, assembly of cohorts, and definition of episodes of care.
Cohort assembly
All eligible providers held a Doctor of Chiropractic (DC) degree. The providers were categorized according to their tier within the ChiroCare Network. By association with providers seen at clinical encounters, patient subjects were assembled into two mutually exclusive cohorts: CC Network (Cohort 1), and CoE (Cohort 2). There were no differences between the two cohorts in the types of services covered or in re-imbursement rates of private and public insurance plans. No patient crossover between provider categories was observed in the data.
Statistical analysis
All statistical analyses were computed using IBM Statistics Software SPSS (Version 23) [12] and all data management procedures followed IRB guidelines. Claims data were examined using the procedure codes as defined by Current Procedural Terminology (CPT) Code. Procedures were categorized as evaluation and management, manipulative therapy, physiological therapeutics, acupuncture, and imaging. The physiological therapeutics category was divided into active (e.g. therapeutic exercise, therapeutic activities, and neuromuscular re-education) and passive (e.g. electric muscle stimulation, ultrasound, massage therapy, manual therapy, and traction). Claims for manipulative therapy of extremities (e.g., 98,943; pertaining to the wrist or knee) were removed from the analyses because such procedures are typically not directly related to treatment of low back pain. Thus, only chiropractic treatments for low back pain were included in the analysis; treatments provided by other types of practitioners were not included. Additionally, because all providers in the current study were chiropractors, CPT 97140 was likely used for manual therapies other than manipulation.
We generated descriptive statistics including means and frequencies on patient age, gender, and other clinical factors (health plan, health plan benefit packages, provider region, and primary diagnosis categories). We categorized health plan benefit packages as either public (benefit packages structured for Medicare or Medicaid populations where costs are at least in part funded by State or Federal government entities) or private (benefits structured for Commercial group or individual populations where costs are privately funded by an employer or an individual). Provider Region was dichotomized as Urban versus Rural). Urban regions are comprised of larger metropolitan areas, while rural regions are comprised of smaller towns or cities. Primary diagnosis of low back pain was categorized as biomechanical lesions, sprains, strains, or other dorsopathies.
Generalized linear modeling (GLM) (with gamma distribution and a log link function) was used to estimate the influence of patient and clinician characteristics on costs, as measured by allowed amounts. Studies examining cost data have demonstrated that the GLM model with the gamma distribution is an optimal choice when modeling skewed cost data [13, 14]. Variables with a p-value p < .05 were considered statistically significant.