The study protocols were approved by the Institutional Review Boards of participating medical centers, and all patients provided written informed consent before any study-related procedures were performed. Patients in the 3 trials were required to have a diagnosis of breast cancer, non-small cell lung cancer (NSCLC; stage IIIb or IV), or gynecologic carcinoma of the ovary, cervix, or uterus. Additional inclusion criteria were the following: ≥ 18 years old, anemic (Hb <11 g/dL at screening), expecting to receive ≥ 8 additional weeks of chemotherapy, a Karnofsky performance scale score ≥ 50%; adequate renal function (serum creatinine concentration ≤2.0 mg/dL), adequate liver function (aspartate aminotransferase or alanine aminotransferase ≤ 2 times the upper limit of the normal range or serum bilirubin ≤ 1.5 times the upper limit of the normal range), and able to complete questionnaires. Patients were excluded from the trials if they had received a red blood cell transfusion within 4 weeks of screening, or erythropoietic therapy within 2 weeks of randomization; had inadequate iron stores (transferrin saturation < 15% and ferritin < 10 ng/L); known positive antibody response to any erythropoietic agent; known history of pure red cell aplasia, of anemia due to hematologic disorders other than chemotherapy-induced anemia, or of uncontrolled hypertension.

The components of the PSQ-An were drawn from other patient satisfaction instruments 26 for other injection treatments (eg, insulin injections, growth factor injections) 2728. Questions from these components were selected as candidate questions for the PSQ-An if they could be modified to capture 1 of the 4 preselected domains of patient satisfaction for anemia treatment (general patient satisfaction, convenience of treatment for the patient and their family and friends, patients' pain and discomfort, and financial burden for the patient). These domains were first identified from the literature review and then selected by the study team based on their relevance to anemia treatment. Redundant questions thought to be capturing the same information as other questions were removed to decrease the size of the final questionnaire. The result was a provisional 21-item instrument comprising 2 parts: a descriptive part (11 items), which included questions about resources devoted to treatment, and a scale part (10 items), which included questions about treatment burden and overall satisfaction. The questionnaire is presented in Appendix (see Additional file 1).

The study sample consisted of 312 adult, English-speaking patients participating in 3 randomized, multicenter trials. This sample size ensures a precision of <5% in the standard errors assuming treatment compliance rates of better than 70%. For logistical and administrative reasons, 3 identical but separate protocols were used, 1 for each tumor type (breast cancer, non-small cell lung cancer, gynecologic carcinoma) with a preplanned analysis of all individual data across studies prespecified in each protocol. Patients with breast cancer (n = 141), gynecological malignancies (n = 69), or non-small cell lung cancer (n = 102) were enrolled in the study and were randomized to treatment with darbepoetin alfa or epoetin alfa for anemia due to chemotherapy. The inclusion of patients with 3 different tumor types reduces the likelihood that treatment satisfaction responses are unique to a single population of cancer patients. Following a 1-week screening period, complete blood counts (including Hb) were measured every 2 weeks prior to dosing. In addition, the 4-week recall patient satisfaction questionnaire was administered at weeks 5, 9, 13, and 17.

Karnofsky Performance Status ratings (0% – Dead to 100% – Normal, no complaints, no evidence of disease) were collected in the trial and converted to Eastern Cooperative Oncology Group (ECOG) Performance Status Ratings to reduce the number of categories with very small numbers of patients used in this analysis. The ECOG Performance Status Rating measures how cancer affects the daily living abilities of the patient 29. The scale ranges from 0 (fully active, no restrictions) to 5 (dead), where lower scores represent better mobility.

The 1-item self-report Global Health question from the MOS SF-36 was included ("In general, would you say your health is...?") with a 5-point Likert Scale, where a higher score represented better health.

The items in the scale part of the instrument originally had response values of 0 to 4 (not at all/mildly/somewhat/moderately/extremely). Values of the 7 negatively stated questions (items 1, 3, 4, 5, 6, 7, and 8) were reverse scored, so that higher values indicate more positive satisfaction. Items were considered for deletion if they met any of the following 3 criteria: 1) missing responses greater than 10% (endorsement rate); 2) more than 50% of participants reporting either the highest or lowest score available (floor/ceiling effect); or 3) significant item-item correlations ≥ 0.70 30. All analyses were completed for data collected at week 5 (test-retest analyses also included data collected at week 9 to maximize available sample size).

Principal component analysis was used to identify meaningful and interpretable factors. The number of factors to retain was based on eigenvalues ≥ 1, with factor loadings serving as an indicator of the degree to which each item was associated with each factor. Items were retained in a given factor if they had a factor loading ≥ 0.40. Multi-trait scaling was carried out to evaluate item convergence within scales and item discrimination across scales. A priori instrument reliability criteria included: 1) item correlation ≥ 0.40 with the total questionnaire (ie, item-internal consistency) 3031, and 2) Cronbach's alpha coefficients = 0.70 (internal consistency) 32.

Test-retest reliability or reproducibility was assessed using the intraclass correlation coefficients (ICC) 33. Responses to the MOS SF-36 Global Health question were used to identify participants with stable health status (ie, whose responses did not change across weeks 5 through 9). ICCs were computed based on this subsample, for the 5 subscale scores at both time points. A predetermined threshold for test-retest reliability was defined as an ICC of 0.70 or greater 3234.

Convergent and divergent validity were examined by estimating Pearson's correlation coefficient and Spearman's rank-order correlation coefficient, between subscales of the PSQ-An and the MOS SF-36 Global Health, Hb level, and measures of time devoted to treatment hypothesized to assess either similar or different constructs 353637. We hypothesized that the scores for the subscales of the PSQ-An measuring aspects of treatment burden would correlate more strongly with the questions relating to time devoted to treatment. Further, the satisfaction subscale of the PSQ-An was expected to have a larger correlation coefficient with the MOS SF-36 Global Health score and Hb level than with resources devoted to treatment. The above correlation coefficients with the MOS SF-36 Global Health score and Hb levels were hypothesized to be positive and those with measures of time required for treatment were expected to be negative.

Discriminant validity was assessed by relating PSQ-An subscale factor scores to 3 variables measuring different aspects of patient health: MOS SF-36 Global Health score, Hb level, and ECOG scores. Mean scores on the subscales of the PSQ-An were compared across response categories of the 3 known measures using analyses of variance (ANOVA) 38. Responsiveness was evaluated by 1) week 5 to 9 effect sizes and 2) ANOVA on week 5 to 9 changes in each PSQ-An subscale and changes in MOS SF-36 Global Health responses. Weeks 5 and 9 were chosen as the best balance between adequate sample size due to patient drop-out in the study and time required for anemia treatments to be effective.

Statistical analyses were performed using SAS version 8.2 for UNIX (SAS Institute, Cary, NC).

The flow of patients initially enrolled in the study is depicted in Figure 1. Note that by week 9, only 80% of the patients remained in the study. Of the 284 patients enrolled at week 5, 264 (92.9%) participants completed the questionnaires at week 5. Table 1 summarizes the demographic and clinical characteristics of the study group classified by anemia treatment. Due to inclusion of breast and gynecological cancers, the sample was primarily female (84.2%, Table 1). Mean ages were 58.7 (11.5) and 61.7 (12.1) respectively in the darbepoetin alfa and epoetin alfa treatment groups. A total of 84.2% percent of the participants were non-Hispanic Whites. Nearly half of all participants (48.4%) had Stage IV cancer.

Response rates on all PSQ-An question items were greater than 90% (Table 2). Mean, standard deviation, and proportion reporting extreme values for each item are also reported in Table 2. As can be seen, 3 items exhibited ceiling effects with over 50% of responses at the highest score (difficulty receiving injection, financial burden, and likelihood of recommendation). Because the highest response marker for each item represented highest satisfaction and effectively captured potential dissatisfaction, we retained these items for further factor development.

Two item-item pairs exhibited correlation coefficients near or greater than 0.70 (Table 3). Interference with daily activities due to treatment-related travel and overall satisfaction were correlated at r = 0.68 (P < 0.001). Overall satisfaction and likelihood of recommending treatment had r = 0.77 (P < 0.001). All 3 items were retained since the correlations were at or just above the predetermined threshold for consideration and we felt each item measured distinct aspects of treatment burden and satisfaction.

Two subscales were identified by factor analysis (Table 4). The first factor was composed of 7 questions that measure burden of treatment, inconvenience, and physical pain. The second factor contained 2 items capturing overall satisfaction. Internal consistency for each subscale is 0.83 (P < 0.001). Test-retest reliability was moderate between weeks 5 and 9 in patients with stable MOS SF-36 Global Health scores (128/284 participants). The item ("relating to difficulty in receiving every injection") did not load strongly on either subscale and was excluded from further analyses.

The Inconvenience subscale correlated negatively with descriptive questions on resources devoted to treatment (Table 5, r = -0.19 to -0.61). The Satisfaction subscale correlated positively with MOS SF-36 Global Health (r = 0.13 to 0.25) and negatively with resources devoted to treatment (r = -0.22 to -0.28). These relationships were largely as hypothesized. However, neither subscale correlated significantly with Hb levels at either week 5 or 9. These results are consistent using both Pearson's correlation coefficients and Spearman's rank-based correlation coefficients.

The 2 PSQ-An subscales correlated moderately with MOS SF-36 Global Health score, Hb level, and ECOG scores (Table 6). ANOVA showed that only the Satisfaction subscale had a significant (P = 0.006) relationship with Global Health.

The effect size between week 5 and week 9 for the Satisfaction subscale was 0.44 (Table 7). This is a moderately large detectable change over this period. In contrast, the effect size for the Inconvenience subscale was a moderate 0.13. Changes over this period in the Satisfaction subscale also correlated with changes in MOS SF-36 Global Health over this period (Table 8). The trends in changes in the Inconvenience subscale scores and in changes in MOS SF-36 Global Health were not statistically significant.