Sample and procedure
A total of 115 women with endometriosis were consecutively recruited from a tertiary referral pelvic pain clinic between April 2003 and September 2006. The Institutional Review Board of the University of North Carolina (UNC) approved the study. All English-speaking and literate women between the ages of 18 and 65 who were new referrals for evaluation of endometriosis-associated chronic pelvic pain at the UNC pelvic pain clinic were eligible for participation. We did not exclude anyone based on pre-existing psychological and/or physical condition. These patients were part of a larger, previously published study on psychosocial correlates of chronic pelvic pain (Meltzer-Brody et al., 2007
). From the larger study (n
= 628 enrolled of 722 eligible), 459 were new referrals to the clinic and were asked to be in a longitudinal study. Of those, 310 agreed to participate and returned at least one follow-up questionnaire (67.5% return rate). For the current paper, we studied only those new longitudinal patients with endometriosis-associated chronic pelvic pain (n
= 123). We excluded eight patients for whom we could not identify medical record data; thus, our cohort consists of 115 participants. Later, we describe in detail our methodology for ascertainment of endometriosis-associated chronic pelvic pain case status.
Following informed consent and prior to examination, participants completed a battery of questionnaires assessing severity of pain, quality of life and psychosocial variables such as catastrophizing. In addition, participants completed a baseline demographic and medical history questionnaire before undergoing standard evaluation by the clinical team. The examining physicians, regionally recognized as experts in the field of pelvic pain, did not review participants’ questionnaire responses. Their findings, diagnostic impressions and recommended treatments [e.g. medications, surgery, referral to physical therapy (PT)] at baseline and over the follow-up year were recorded in electronic medical records. A team of four trained research associates then abstracted these data.
We used two complimentary methods for ascertaining case status as endometriosis- associated pelvic pain: (i) patient self-report on entry per demographic questionnaire and (ii) review of electronic medical records. Based on our record review, we could confirm a surgically based diagnosis of endometriosis for 111 patients; four women were surgically diagnosed during the study period. Therefore, our cohort consists of 115 women who were surgically diagnosed with endometriosis and experienced intractable pelvic pain despite conventional treatment modalities; 93% completed the 1 year pain questionnaire.
Demographic and clinical history
The baseline study entry questionnaire provided demographic and medical history information concerning age, years of education, race, marital status, duration of pelvic pain, number of prior surgeries for pelvic pain and previous hysterectomy. Number of prior surgeries for pelvic pain was truncated to six surgeries in order to reduce outliers. As a result of our population demographics, the race and marital status variables were treated as bivariates in data analyses (e.g. Caucasian versus other, married versus not married). When self-reported data were missing for the duration of pelvic pain and previous hysterectomy variables, information on these variables was obtained from the medical records.
Medical records were abstracted for additional demographic and clinical variables (e.g. parity, co-morbid pain disorder) as well as treatments (e.g. surgical versus medical) utilized over the study period. For secondary analyses, we examined those having the most common type of surgery (e.g. hysterectomy) versus other types of pelvic surgery (e.g. oophorectomy, lysis of adhesions). We also abstracted information about the three common types of medical treatments used before the study and during the follow-up year: (i) PT, (ii) hormonal therapies (e.g. oral contraceptive pills, leuprolide, levonorgestrel-releasing intrauterine system) and (iii) central nervous system medications (e.g. antidepressants, anticonvulsants, sleep aids, muscle relaxants). Patients were considered to have another co-morbid pain disorder if any of the following were noted in the electronic medical records from the initial visit: chronic headaches (migraines), irritable bowel syndrome, interstitial cystitis, fibromyalgia, chronic fatigue syndrome, temporomandibular disorder or chronic lower back pain.
To determine the severity of pelvic pain at baseline and follow-up, the short form McGill pain questionnaire (SF-MPQ) was used (Melzack, 1987
). The SF-MPQ has two components, the present pain intensity (PPI) and pain rating index (PRI). The PPI component of the questionnaire consists of a visual analog scale and is conventionally used to assess the present pain levels. Consistent with conventional practice, we used the PRI component (McGill total pelvic pain score) in this study since it provides a global multi-dimensional pain severity rating over time. The McGill total pelvic pain score (PRI component) is composed of 15 groups of qualitative descriptors (11 sensory and 4 affective) that are rated on an intensity scale (0 = none, 1 = mild, 2 = moderate and 3 = severe). Participants were asked to rate the descriptors of the pelvic pain they had experienced in the previous 2 weeks. The McGill total pain score was calculated from the sum of the 15 items, with a range of 0–45. The SF-MPQ is highly correlated with the longer version of the MPQ (Katz et al., 1999
). It has demonstrated a high test-rest reliability of 0.96 (Grafton et al., 2005
) and a high internal consistency with a Cronbach's α
of 0.86 (Mason et al., 2008
). Cronbach's α
was 0.90 at baseline in the current study.
To assess mental health, we used the mental health subscale (norm-based scoring) of the short form 12 Health Survey (SF-12) where a high score indicates better mental health. The SF-12 has a high internal consistency and reproducibility with a Cronbach's α
of 0.87 (Ware et al., 2002
). It has been shown to be able to discriminate between types and severity of disease between patients with medical disorders alone as well as those with both medical and psychological disorders (McHorney et al., 1993
). The mental health subscale has strong predictive properties for health outcomes with a test-retest reliability of 0.76 (Ware et al., 1996
). SF-12 Cronbach's α
was 0.89 at baseline in the current study.
To measure catastrophizing of pain, we used the catastrophizing subscale of the Coping Strategies Questionnaire: CSQ (Rosenstiel and Keefe, 1983
). The CSQ was designed to evaluate cognitive and behavioral coping strategies in chronic pain syndromes. The catastrophizing scale ranges from 0 to 6, with a high score indicating higher catastrophizing (e.g. ‘I feel I can't stand it anymore’; ‘It is terrible and I feel it's never going to get any better’). This scale has been shown to predict poor health in other clinical samples, such as Irritable Bowel Syndrome patients (Drossman et al., 2000
). Data analyses were performed with the continuous measure; however, in the text, we report the following groups based on their catastrophizing score (range 0–6): mild (1–2), moderate (3–4) and severe (5–6). The catastrophizing subscale of the CSQ has extensively been utilized in the literature. It has high internal consistency and reliability with a Cronbach's α
of 0.78 (Rosenstiel and Keefe, 1983
). This scale has been found to have strong predictive properties for health outcomes among patients with painful conditions (Robinson et al., 1997
; Schanberg et al., 1997
) witha test-retest reliability varying from 0.80 to 0.91 (Rosenstiel and Keefe, 1983
). Cronbach's α
was 0.73 at baseline in the current study.
Statistical analyses were performed using SAS 9.1. Descriptive statistics (mean and standard deviation for continuous variables; frequency and percentages for bivariate variables) for demographics, pain report and psychological variables are reported for the entire cohort in the ‘Descriptive Analyses’ section of the results below. We examined the distributions of the pain variables at baseline and at 12 months to ensure approximate normality. Continuous variables (e.g. catastrophizing) were analyzed as continuous variables but were grouped into categories of mild, moderate and severe for presentation and clarity purposes only. Paired Student's t-tests were used to compare changes in pain over time for the group as a whole.
In order to determine predictors for pain severity at baseline and at 12 months, we ran multiple regression analyses allowing variables with P < 0.10 or those with clinical interest to stay in the models (Table ). For regression analyses of baseline pain (McGill total pelvic pain), we tested the significance of the following predictor variables: (i) demographic factors (age, education, race, marital status, parity), (ii) previous medical history (duration of pelvic pain, number of surgeries for pelvic pain, PT for pain) and (iii) baseline mental health (catastrophizing, SF-12 mental health).
Factors associated with baseline pain severitya.
For regression analyses of McGill total pelvic pain at 1 year, we first entered baseline pain into the regression models in order to assess the change in pain over time. We then investigated the joint and independent contributions of baseline variables and treatments used during the 1-year follow-up period. Thus, the following variables were included in the predictive models of total pelvic pain at 1 year: (i) baseline McGill total pelvic pain score, (ii) baseline mental health (catastrophizing, SF-12 mental health), (iii) demographic factors (age, race, marital status, education) and (iv) interim treatments (surgery, PT, medications) (Table ). These regression models were run with the continuous form of predictor variables (e.g. catastrophizing, education, age, number of prior pelvic surgeries). For presentation purposes though, we also ran post hoc general linear models with least-squared means (Student's t-tests) for categorical predictors in order to determine which categories of patients (e.g. severe catastrophizing, less than college degree) had more pain. To reduce the number of variables (maximum of nine in all our analysis), we first tested demographic and medical history variables and only included the significant ones in the final model. In addition, due to co-linearity, mental health variables (SF-12 and catastrophizing) were run separately.
Predictors of pain severity at 1 year follow-upa.
Our cohort primarily consisted of Caucasian (85%), married women (67%), aged 19–45 (85%) with at least some college education (70%); 46% were parous. With a mean duration of 5.6 (±5.3) years of pelvic pain, the majority of our subjects had underwent two previous surgeries (mean 1.9 ± 1.7) for treatment of pelvic pain; 34% (n = 39) had a previous hysterectomy at entry into the study. Slightly over half of our participants (n = 64; 55.6%) had been diagnosed with at least one other co-morbid pain condition (e.g. fibromyalgia, migraine headache).
Compared with women in the US population, women in our study were considerably below average (approximate 25th percentile) in their mental health status, with a mean SF-12 score of 42.8 ± 10.3 (Ware et al., 2002
). Overall, 43.47% of participants scored moderate to severe on catastrophizing (Fig. ). Most experienced moderate levels of pain (mean McGill pain score 19.45 ± 10.0), with scores higher than those reported for fibromyalgia (15 ± 6.7) patients (Geisser et al., 2003
Differences between baseline and 1-year McGill total pelvic pain within catastrophizing groups (P-values are from paired Student's t-tests between baseline and 1-year McGill total pelvic pain scores within each catastrophizing group).
Most demographic, clinical and psychosocial characteristics did not differ between patients undergoing medical therapy or medical in conjunction with surgical therapy. However, on average, the non-surgical group entered the study having a longer duration of pelvic pain by 2.4 years (P = 0.01), one additional past surgery for pelvic pain (P = 0.05) and a greater likelihood of having a previous hysterectomy (P = 0.008).
Most patients improved in their pain scores from entry to follow-up (74.8%). Patients reported an average of a 37.4% decrease in pain (P < 0.001) during the year. Both surgical and non-surgical patients improved significantly over time (P < 0.001), with a trend for those having surgery to have a greater mean pain improvement (50.3%) compared with those without surgery (31.2%; P = 0.05). The most common medical treatments included hormonal therapies (59.1%) and central nervous system medications (34.8%). The most common surgeries performed over the year included hysterectomy (11.3%), oophorectomy (11.3%) and lysis of adhesions (10.4%). In further analysis, we found that the trend for improved 1-year pain outcomes among those undergoing surgery was attributable to interval hysterectomy. Thus, in later analyses, we focused on hysterectomy rather than all surgeries.
Baseline pain analyses
We first analyzed the demographic, previous medical history and mental health correlates of baseline total pelvic pain. Greater pain at study entry was associated with younger age (P = 0.02), less education (P = 0.01), more previous pelvic surgeries (P = 0.05), past PT treatment (P = 0.03) and coping with pain by catastrophizing (P < 0.001). Specifically, further analysis showed that those 30 years or younger, those having less than a college degree, those having three or more pelvic surgeries and those who cope via severe catastrophizing had the worst pain at baseline. These predictors accounted for ~35% of the variance in baseline pain scores. Catastrophizing alone in these models explained 21% of the variance in pain scores, with those coping by severe catastrophizing having about a 9-point higher pain score (range 0–45) than those who cope by mild catastrophizing. Since the mental health (SF-12) and catastrophizing measures were highly correlated (r = −0.53; P < 0.001), they could not be in the statistical model simultaneously due to co-linearity. Good mental health (i.e. high SF-12 mental health score) was also associated with less pain at baseline when catastrophizing was not included in the model (β = −0.25; P < 0.001).
Pain at 1 year
Controlling for baseline pain, we examined the demographic, medical history, mental health and interval treatments predicting pelvic pain at 1 year (Table ). Those who had higher pain scores at 1 year after controlling for entry pain were nulliparous (P = 0.002), less likely to have a hysterectomy during the follow-up period (P = 0.008), more likely to receive PT during the follow-up (P = 0.04), slightly more likely to be less educated (P = 0.07) and more likely to cope via catastrophizing at study entry (P = 0.04). Having a hysterectomy explained ~8% of the variance in pain at 1 year, after controlling for baseline pain. Of the 13 people who had a hysterectomy during follow-up, only one reported follow-up pain above the median pain score for the entire cohort, compared with 59% of those who did not have this surgery.
The predictors of pain at 1 year did not change when we excluded those subjects who entered the study with a previous hysterectomy. Specifically, the standardized parameter estimate and significance (β = −0.287; P = 0.01) for the effect of interim hysterectomy were still notable when those with a prior hysterectomy were excluded from the analyses, compared with the results for the entire cohort shown in Table .
Catastrophizing was a significant predictor of pain at 1 year, explaining 3% of the variance in pain outcomes. At baseline, however, catastrophizing accounted for 21% of the variance in pain levels. This seemingly unexpected observation is attributable to the inclusion of baseline pain in the 1 year statistical model. Since baseline pain explained a large portion of the variance (41%) in 1 year pain outcomes, not much variance was left to be accounted for by catastrophizing despite its significant association with 1 year pain levels. Changes in pain scores within and between catastrophizing groups (mild, moderate, severe) are shown in Fig. . Although all groups tended to decrease in their pain scores, those who exhibited severe catastrophizing at baseline started and ended the study with higher pain levels. Only those with mild (P < 0.001) and moderate (P < 0.001) levels of catastrophizing showed significant pain improvement at 1 year.