The methods for this project were modeled after the RAID score methodology but with some modifications based on the differences between the diseases.
The RAID score
An international group of ten patients with rheumatoid arthritis (RA) from ten countries was selected in the first phase (17 dimensions) where the disease had important impact. The second step was performed by 100 patients from the same countries, who ranked the dimensions for importance to reduce the number of dimensions to seven [11].
In the second phase, 500 patients from the same countries were asked to distribute 100 points between the 7 dimensions. Weights were ranked within each individual, and the average ranks were used for the final weight of each dimension. These ranks formed the basis for the final weights and were linearly transformed to a 0–100 scale. These weights were multiplied by the results of the numeric rating scales (NRS) and added together into the RAID score with a range from 0 to 10 [12].
An international multicentre cross-sectional and longitudinal study of consecutive RA patients from 12 European countries was conducted in the third phase to examine the psychometric properties of the combinations of instruments that might be included within the RAID and the use of NRS for each dimension. Construct validity was assessed cross-sectionally.
The PACADI score
First phase: selection of dimensions and follow-up examinations
Different from the patients with RA, who were familiar with their disease, the patients with PC only recently faced a severe disease at time of inclusion. For the PACADI score, based on an ethical consideration, individual settings were chosen to replace the group format. The study is a single center study on a tertiary level.
The goal was to include 30 patients for the identification of dimensions. There is no definitive answer to decide the sample size in qualitative approaches. However, the principle of saturation, i.e., to reveal the full range of important perceptions, is regarded as an indicator. A critical review from Yamazaki et al. found a median sample size of 36 [16]. The first sample consisted of 52 consecutively referred patients between November 2008 and July 2009 with symptoms and findings indicating PC. After the diagnostic procedures, 41 had confirmed diagnoses of PC (ICD10 C25*) based on cytology or histology in 92.7 % of the patients. Three patients (7.3 %), in whom all clinical, biochemical, and radiological data were consistent with pancreatic adenocarcinoma, were included based on findings by imaging modalities. Demographic and disease characteristics at baseline of the 41 patients with confirmed PC are shown in Table 1 (for the entire enrolled population (n = 52) see online supplementary Table S1).
Table 1 Demographic and disease characteristics at baseline for sample 1 for selection of dimensions and sample 2 for weighting and preliminary validation
With a qualitative approach, the patients were asked an open-ended question to identify up to 10 important dimensions of health related to the impact of the disease. The patients were shown a list of 56 dimensions (Table S2) from frequently used and relevant generic and disease-specific PRO instruments, and were then asked to report as free text the most important dimensions according to their personal opinion, but not in prioritized order. The selected dimensions were given priority according to importance by the patients immediately after selection, giving 1 to the most important, 2 to the next, etc.
Age, gender, body weight, and height at baseline were recorded. Body mass index (BMI) was computed. Regular weight prior to disease onset was recorded, and weight loss to baseline was calculated. Date of surgery was registered, if performed.
The patients also completed Edmonton Symptom System (ESAS) [4] and EuroQol-5D (EQ-5D) [6]. ESAS consists of ten NRS (scales 0–10). The Norwegian ESAS version was used with the following ten items: pain at rest, pain at movement, fatigue, nausea, dyspnea, dry mouth, loss of appetite, anxiety, depression, and sense of well-being [10]. EQ-5D is a five-item instrument that can be used in cost–utility analyses. Three of the items concern physical function, and the two last items address anxiety and pain. Each question has three response options (none/minor, moderate, or major problem). A single score from zero to one can be calculated.
We repeated the selection of important dimensions of health, after 1 and 2 months in a longitudinal design, to examine if the selection of important dimensions changed during the disease course. Data were available in 29 out of the 41 patients. Three patients died before the follow-ups were completed. Three dropped out, and six were too weak and died between 4 and 8 months after inclusion.
Second phase: weighting
Patients with PC (sample 2) were included between September 2009 and September 2011 based on the same inclusion criteria as in sample 1. Out of the 110 patients with suspected PC (online supplementary Table S1), 80 had confirmed PC (Table 1). Selection of important dimensions of health and data collection were similar to sample 1 (demographics, ESAS, and EQ-5D), but included one NRS (scales 0–10) for each of the selected dimension of the PACADI score (online supplementary Table S3, showing the eight NRS in Norwegian and in English translation (see Table S3 for description of the translation process)) [17]. Each NRS addressed the impact on each dimension during the last week. The last week was chosen as a time frame for overall impression of the health dimensions in the PACADI score, since patients often report fluctuating severity of their health problems. The 1-week time frame is similar to the RAID score and different from the ESAS (same day). Time frame varies between instruments, and 1 week has been found as valid as momentary assessment [18].
For the weighting exercise, the following question was asked: “Considering your entire disease duration, please distribute 100 points among the following eight dimensions of health based upon the impact of the disease on these dimensions.” This methodology is identical to the methodology for assigning weights to the dimensions of the RAID score [9] and as described by Ruta et al. [23].
The weights of the PACADI profile were ranked from 8 to 1 in each patient (8 was given to the dimension with highest weight). If two dimensions had identical weights, they were given the same rank. The mean of the rank of weights for each dimension was calculated and was then normalized to a scale from 0 to 1 (i.e., the sum of all weights was equal to 1.0). NRS were also used in the RAID score and were performed, as well as other and more comprehensive tools within the same dimension. In addition, NRS is faster and more feasible to complete for patients than more comprehensive questionnaires and visual analogue scales [19]. The final step in computing the PACADI score was to multiply the NRS values with the normalized final weights and then to provide a sum of these products in each individual patient.
Third phase: preliminary validation
The preliminary validation in sample 2 included a comparison of the PACADI scores across gender, between patients undergoing versus not undergoing surgery, between patients with and without confirmed PC in sample 2, and to examine correlations between the PACADI score and age, ESAS items and EQ-5D.
Statistics
Parametric statistical methods were preferred since data appeared normally distributed based on Q–Q plots, and mean and median values were similar. Parametric methods were performed robustly [20] and were used in the development of the RAID score. Demographic continuous variables as well as PACADI score, ESAS and EQ-5D are presented as mean (SD) values, and associations between PACADI and continuous demographic variables, ESAS and EQ-5D were examined with Pearson’s correlation coefficient. Two-sample t tests were used for the comparison of the PACADI score across dichotomized variables. All analyses were performed in SPSS version 18. Two-tailed p values <0.05 were considered statistically significant.
Regulatory and ethical aspects
All data were stored with a numeric identifier in a separate, secured research database detached from the identifier code list. The code list was properly stored with accessibility for the principal researchers only. The regional ethical committee approval of the protocol and patient information/consent form was obtained separately for phases 1 and 2. The privacy protection supervisor evaluated and accepted the data-handling procedures. All participating patients signed a written informed consent.