Study design and setting
This retrospective, practice-based research study of cardiovascular inpatients was conducted at ANW, a 630-bed teaching and specialty hospital in Minneapolis, MN. The Penny George Institute for Health and Healing (PGIHH) at ANW was founded in 2003 and offers hospitalized patients, through electronic physician and nurse referrals, a wide-array of integrative health services at no charge to patients [23]. All IM practitioners at PGIHH are formally trained and have necessary licensure and/or certification in their area of specialty (e.g., aromatherapy, acupuncture, massage, music). Referral criteria include: (a) the patient is able to participate in integrative health intervention, and (b) patient concerns include pain, anxiety/stress, elimination problems, nausea/vomiting, insomnia, coping with change in health/well-being, or maintaining/prolonging a pregnancy.
Study population
All cardiovascular inpatients age 18 years or older at ANW, who were admitted between July 1, 2009 and December 31, 2012, were included in the study population. Patients seen as outpatients, in the emergency room, and who were in the hospital solely for observation were excluded. Medical record data were obtained on all eligible inpatients and cardiovascular patients were retrospectively identified. All patients whose medical record data were obtained gave written permission upon hospital admission to use their records for research purposes.
The study population included those with diseases of the circulatory system, identified using the International Classification of Diseases (ICD), 9th Revision, Clinical Modification diagnosis codes (390-459). Any admission that had at least one of these ICD-9 codes as the admission’s primary or secondary diagnosis or any hospital encounter-level diagnosis was eligible for the study.
We created non-mutually exclusive indicators pertaining to five circulatory system diseases: diseases of arteries, arterioles and capillaries (440-448); cerebrovascular disease (430-438); hypertensive disease (401-405); ischemic heart disease (410-414); and diseases of pulmonary circulation (415-417). Patients of all other circulatory system diseases were grouped into an ‘other’ category.
The study was approved by the Institutional Review Board of Allina Health with a waiver of informed consent.
Measurements
Demographic and admission characteristics
Data extracted from medical records included patient age at time of admission, sex, race, marital status, and health insurance status. The data included the All Patient Refined Diagnostic Related Groups (APR-DRG) [24] severity of illness measures calculated from patients’ diagnoses codes. The measure includes four categories of severity: 1) minor, 2) moderate, 3) major, and 4) extreme. Data pertaining to each IM session were routinely documented in a customized documentation flowsheet within the medical records.
Integrative medicine therapies
IM practitioners used their clinical judgment to provide therapies, within their scope of practice, they deemed necessary and therapeutic for each patient, after consulting with the patient. Many patients received IM therapy numerous times throughout their hospital admission. The term ‘session’ is used to define each unique administration of IM therapy, distinguished by time of procedure, within a hospital admission. For the present analyses, IM therapies were placed into one of three broad categories: bodywork (BW), which included craniosacral therapy, medical massage, and reflexology; mind-body and energy therapies (MBE), which was divided into separate mind-body and energy subcategories; and traditional Chinese medicine (TCM), which included acupressure, acupuncture, and Korean hand therapy. Importantly, patients could receive therapy from more than one category during each session, which has been defined as combination therapies. The presence or absence of each of these IM therapies was coded at each session such that BW, MBE, TCM, and any combination of these therapies were mutually exclusive.
Pain and anxiety scores
IM practitioners collected patients’ self-reported pain and anxiety scores directly prior to and after each IM session. Practitioners requested patients to provide a single number to indicate the level of pain they were currently experiencing on an 11-point numeric rating scale where 0 was defined as ‘no pain’ and 10 was defined as ‘worst pain imaginable’. Similarly, practitioners recorded anxiety scores using the same methodology, where 0 was ‘no anxiety’ and 10 was ‘worst anxiety imaginable’. The primary endpoints were changes in pain and anxiety scores, calculated by subtracting the pre-score from the post-score. Zero to 10 numeric rating scales for pain have been found valid and reliable [25, 26].
Analytic data set
A total of 57,444 cardiology-related hospital admissions were identified from medical records. During data cleaning, 149 hospital admissions were removed due to missing demographic data (51 admissions) or inability to determine severity of illness (98 admissions), resulting in 57,295 cardiology admissions from 37,259 unique patients. Of the 57,295 admissions, 6,589 (11.5%) had 16,344 IM therapy sessions (average of 2.48 per admission). In many cases, practitioners were unable to collect pre- or post-pain and anxiety scores or patients reported no pain or anxiety. Only patients who reported pre- and post-pain scores and/or pre- and post-anxiety scores, and pre-pain/pre-anxiety scores greater than zero, were included in the subsequent analyses examining changes in pain and anxiety after receiving IM therapy.
Because IM therapies were observed at the hospital admission level, but pain and anxiety scores were assessed at the IM session level, one session was randomly selected from each remaining hospital admission in order to keep the level of analysis consistent between the selection and score change equations (see below). Thus, we dropped all hospital admissions with only missing scores or pre-pain or -anxiety scores equal to zero. This method produced a sample of 54,163 hospital admissions for the pain model, of which 3,457 (6%) had IM therapy, and 52,572 admissions for the anxiety model, of which 1,866 (4%) had IM therapy.
Statistical analysis
IM therapy utilization
Logistic regression was used to predict the probability of receiving any IM therapy during a hospital admission as a function of patient demographics, circulatory system disease diagnosis, severity, and health insurance status, and odds ratios for each covariate are presented. A p-value of less than 0.05 was used to signify statistical significance. We used a random sample of 25,000 observations to test the goodness-of-fit for our model using the Hosmer-Lemeshow test [27]. We did not use the full sample because the Hosmer-Lemeshow test has been shown to likely reject the null hypothesis of a good fit even for models that fit well when the sample size is greater than 25,000 due to increased statistical power [28]. The percent of admissions correctly classified by the model were also calculated.
Pain and anxiety
First, to determine if IM therapies were associated with reductions in pain and anxiety, paired t-tests were conducted using the null hypothesis that the pre- and post-pain or anxiety scores were equal.
Next, multivariate regression was used to estimate reductions in pain and anxiety during IM sessions. Because patients receiving IM therapy may systematically differ from the general sample of cardiovascular patients, an ordinary least squares model could produce bias parameters when generalizing results. To address this bias, a Heckman selection model [29] was used to account for selection into the sample of IM therapy recipients.
To correctly identify the parameters that affect pain and anxiety, at least one variable in the selection-equation (i.e. utilization of IM therapy) should be specified which predicts IM therapy use, but does not affect changes in pain or anxiety. Since marital status and health insurance status were expected to fit this criterion, our model predicted selection into the sample of IM sessions using all patient demographic, circulatory system disease diagnosis, severity, and health insurance variables (the same set of covariates as our logistic regression predicting IM therapy use). Changes in pain and anxiety scores were estimated using diagnosis, age, sex, race, severity, and the inverse Mills ratio calculated from the selection-equation to control for selection. Additionally, we estimated a second model, which included IM therapy categories, to determine if differential effects between the categories existed.
All analyses were conducted in Stata Version 13 (StataCorp LP; College Station, TX).