Patients
Patients consulting a physical therapist for a complaint of neck pain from January 1 2012–June 30, 2013 who were continuously insured under one plan, University of Utah Health Plans (UUHP), were eligible to be included in the analysis. Patients insured under UUHP were participating under a Medicaid managed care plan (a government subsidized plan) or a privately insured, employer-based plan. Patients included in this study sought care from hospital-based or an ambulatory physical therapy clinic in Salt Lake City, Utah and surrounding coverage areas. This study was approved by the University of Utah Institutional Review Board.
We identified patients with a new consultation with a healthcare provider for a diagnosis of neck pain using claims data on the basis of the following International Classification of Diseases-Ninth Revision (ICD-9) codes: 721.0, 721.1, 722.0, 722.4, 722.71, 722.81, 722.91, 723.0–723.9, 739.0, 739.1, and 847.0. We defined the date of the first consultation with a healthcare provider with a neck pain ICD-9 code as the index visit. We only included patients with an ICD-9 diagnosis of neck pain on the index visit who did not have a recorded healthcare encounter in the preceding 90 days, in order to reflect a sample of patients seeking care for a new episode of neck pain. The 90-day washout period was used to provide an adequate amount of time to reflect a pain-free state while acknowledging the biases associated with a washout period less than 1 year [26]. Therefore, we excluded any patients who had a neck pain ICD-9 code associated with any claim in the preceding 90 days from the index visit.
Identifying the timing of physical therapy
We further identified patients who sought care from a physical therapist from billed procedure and revenue codes for physical therapy in the claims data (Additional file 1). To determine the timing of physical therapy consultation, we calculated the number of days between the index visit (first consultation with a healthcare provider) and first physical therapy consultation. We further identified three groups of patients seeking care for a new episode of neck pain. If a patient consulted a physical therapist on the index visit or within 14 days of the index visit, we categorized these patients as receiving “early physical therapy consultation”. If patients consulted a physical therapist within 15–90 days of the index visit, they were categorized as receiving “delayed physical therapy consultation”; if patients consulted a physical therapist after 91 days to any time within the following year from the index visit, they were categorized as receiving “late physical therapy consultation”. The early and delayed physical therapy consultation groups were selected based on previously published literature that described early physical therapy consultation within 14 days and delayed consultation between 14 and 90 days. [27] The late physical therapy consultation group was used as the reference group in analyses.
Comorbidities
We wished to identify comorbidities that may influence physical therapy outcomes, neck pain prognosis or healthcare seeking behaviors from recorded ICD-9 codes in the claims data within the 1-year period following the index date. We recorded the following provider-entered comorbidities: low back pain [21], fibromyalgia [28], chronic or generalized pain [29], substance abuse, depression and anxiety [30], tobacco use and obesity (see Additional file 1 for ICD-9 codes used for co-morbidity identification).
Exclusion criteria
We excluded patients younger than 18 years of age (n = 30) and patients with ICD-9 codes for diagnoses that may adversely affect healthcare costs and utilization such as a diagnosis of a spinal cord injury (n = 3), vertebral fracture (n = 3) or malignant neoplasm (n = 17). ICD-9 codes for comorbidities were recorded any time with in the 1-year time period from the index visit. See Fig. 1 for sample derivation and Additional file 1 for ICD-9 codes for exclusion.
Healthcare process variables
We identified process variables associated with the episode of care for neck pain. We recorded the proportion of patients who were privately insured versus insured under Medicaid managed care with UUHP and we recorded the UUHP concurrent risk score, an indicator of health and cost risk [31]. We calculated the healthcare episode of care (HC-EOC) as the number of days from the index visit to the last provider encounter with a recorded ICD-9 code for neck pain. The duration of physical therapy treatment (PT-DOT) was calculated as the number of days from the first physical therapy encounter to the last recorded physical therapy encounter. We also calculated the number of physical therapy visits during the PT-DOT. Median and Interquartile Range (IQR) was reported with HC-EOC, PT-DOT and for number of physical therapy visits.
Healthcare utilization outcome variables and cost
We wished to explore the association of timing of physical therapy on interventions with conflicting evidence for effectiveness or indication [22,23,24, 32]. We identified healthcare utilization outcomes from billed procedure codes which has an associated ICD-9 neck pain diagnosis for a 1-year period following the index visit (see Additional file 2). Healthcare utilization outcomes were recorded from the index visit and the following 1-year and may have occurred before and after the physical therapy consultation. We identified patients who received cervical spine injections or nerve blocks; imaging of the cervical spine (MRI, CT and X-ray); or were prescribed an opioid, as identified by therapeutic class codes for Opioids in the claims database (H3A, H3H, H3M, H3N, H3U, H3R), within 1 year following the index visit (Additional file 2). We recorded the billed amounts (costs) from the claims database for all procedures, visits and equipment associated with a neck pain related ICD-9 diagnosis in the 1-year following the index visit.
Data analysis
STATA 14.2 was used to conduct all statistical analyses. Baseline characteristics, comorbidities, healthcare process variables and unadjusted healthcare utilization variables were compared between physical therapy consultation groups using one-way ANOVAs for continuous variables and chi-squared tests for categorical variables. When comparing the HC-EOC, PT-DOT and number of physical therapy visits, Kruskal-Wallis tests were used due to violations of assumption of normality [33].
Logistic regression was used to compare the odds of healthcare utilization of injections; imaging (MRI, CT or X-ray); and being prescribed an opioid within 1 year from index visit between physical therapy timing groups. Based on previous literature, the covariates of age [34] and gender [34] were included in each model. Based on univariate comparisons, comorbid low back pain [14] and comorbid chronic pain [6] and comorbid substance abuse were also included in each model. Insurance plan type (private versus Medicaid managed care) and UUHP concurrent risk score were also added to the models to account for health-system factors that may affect utilization outcomes. The late physical therapy consultation group was used as the reference group in all analyses and no interaction terms were included in the models.
Mean 1-year neck pain-related billed healthcare costs were compared between physical therapy consultation groups using generalized linear modeling (GLM) with gamma distribution and log link function. GLM was used to allow for parametric analytic methods while accommodating non-normal distribution of cost in order to make inferences about the mean costs directly [35,36,37]. Both unadjusted and adjusted analyses were performed. Covariates in the adjusted model included age [34], gender [34], comorbid low back pain [14], comorbid chronic pain [6], comorbid substance abuse, UUHP concurrent risk score and insurance plan type (Medicaid or private). No interaction terms were included in the model.