Development and validation of an abbreviated version of the Trust in Oncologist Scale—the Trust in Oncologist Scale–short form (TiOS-SF)

The full-length 18-item Trust in Oncologist Scale asks patients to report their trust in their own oncologist. It consists of the four dimensions of trust: Fidelity, Competence, Honesty and Caring [6], and two global trust items. All items consist of a proposition in the second person singular, with a 5-point Likert answering scale ranging from 1 (strongly disagree) through 5 (strongly agree). Three items are negatively phrased. After recoding the negatively phrased items, a mean score is calculated by dividing the summed scores by 18. Hence, the possible scores range from 1 to 5, with 5 indicating the highest level of patients’ trust in their oncologist. The scale was originally developed in Dutch, and tested for validity and reliability on a sample of Dutch cancer patients (n = 423) [12]. It was later translated to English and validated on a sample of Australian cancer patients (n = 175) [11].

We aimed to create a one-dimensional five-item reliable and valid short form for the TiOS (the TiOS-SF), including one item from each dimension and one global item to ensure content validity. Data of both previous validation studies were used to ensure generalizability of results. Items were selected with high face validity (conceptually central to the dimension from which they were selected) and high statistical performance (high factor loadings, item-scale correlation, high correlations with individual items and maximal variability in responses). We aimed to select items for which results corresponded between the Dutch and Australian data sets. To test structural validity, the resulting five items were subjected to exploratory factor analysis (EFA) using SPSS 22 (IBM Ltd., Chicago, IL), with orthogonal rotation (varimax). An eigenvalue cutoff of >1.0 was used to indicate separate factors.

Adult cancer patients were recruited through PanelCom, a panel including patients currently or previously treated for cancer (see www.​healthcommunicat​ion.​nl/​panelcom/​). A random selection of 148 panel members was approached by e-mail and invited to fill in an online questionnaire within the next week, with the possibility to refuse participation. A reminder was sent to non-responders after 2 weeks. Approval for the study was obtained from a local, certified Medical Ethics Committee. Nevertheless, electronic informed consent was obtained for all participants. To perform test-retest correlations, consenting participants received a second questionnaire 4 weeks later. Consenting patients received the second questionnaire which contained only the five TiOS items.

In addition to the TiOS-SF, participants reported their socio-demographic characteristics, including gender, age, education level and country of origin. Self-reported disease characteristics were time since diagnosis, type of cancer and the number of previous consultations with the present oncologist. Four measures were assessed to establish discriminant validity. First, satisfaction should theoretically be correlated, although not perfectly, with trust scores, which was confirmed in previous psychometric research [4, 7, 12]. Satisfaction with the oncologist was measured with the 5-item Patient Satisfaction Questionnaire (PSQ), assessed on a 100-point visual analogue scale [8]. However, due to technical problems in the online questionnaire, data of four of the five PSQ items was lost. From experience, we know the first item is strongly correlated to the other items of PSQ (r _s > 0.60) [5, 6]. For that reason, and to avoid data loss, the data of the first item instead of the average score on all five PSQ items was used to assess the relation between satisfaction and trust. Three additional items assessed, using 5-point Likert scales, (1) whether patients would recommend their oncologist to their close ones (completely disagree–completely agree), (2) whether patients would get into contact with their oncologist if something would bother them (certainly not–most certainly) and (3) patients’ reported level of trust in the Dutch healthcare system (a lot of trust –very little trust). These measures were expected to correlate substantially, yet not perfectly with trust, based on previous findings [4, 12, 15, 16]. The second questionnaire, aimed to establish temporal stability, included only the TiOS-SF.

Structural validity was determined using EFA with orthogonal rotation (varimax) to allow for other factor solutions than the presumed one-dimensionality. An eigenvalue cutoff of >1.0 was used to indicate separate factors. An indication for the scale’s reliability was assessed by determining internal consistency (item-total and inter-item correlations) and reliability (Cronbach’s α and temporal stability) [14]. High internal consistency was expected and indicated by high item-total correlations r > 0.50 and inter-item correlations of r > 0.50 [17]. Reliability would be indicated by Cronbach’s α > 0.80 [18] and by high temporal stability as indicated by correlations >0.50 between mean and individual item scores on the test and retest.

Convergent validity was assessed using non-parametric correlations between mean TiOS-SF scores with scores for satisfaction (PSQ item), recommending the oncologist, contacting the oncologist and trust in Dutch health care. Moderate, but not perfect correlations (r _s .20–0.50) were expected.

Finally, correlations between TiOS-SF scores and sample characteristics, i.e., age, gender, education, ethnicity and time since diagnosis were explored. All analyses were performed using SPSS 22 (SPSS Statistics version 22; IBM Ltd., Chicago, IL).

Of the 148 patients invited to participate, 128 patients (86%) responded, of whom 101 patients consented to participate (response rate 68%). The most frequently mentioned reason for patients not to participate was that they did not have recent contact with their oncologist. Data of nine respondents were excluded because their data on the TiOS-SF were missing. In total, 92 questionnaires were included in analyses for the test and 61 for the retest (response rate retest 70%).

Sample descriptives are presented in Table 2. Mean trust in the oncologist was 4.30 (SD = 0.98, range 1–5, skewness − 1.95 (SE = 0.25)). Skewness was below 2 for all individual items at test as well as retest. Nevertheless, substantial skewness did occur, indicating ceiling effects for trust scores.

EFA on the TiOS-SF indicated a one-factor solution. This factor, with eigenvalue 4.09, explained 82% of the total variance. Factor loadings are displayed in Table 3.

High reliability of the shortened TiOS was indicated by a Cronbach’s α = 0.94. Initial test-retest correlations indicated poor temporal stability of the TiOS-SF (r = .35, p < 0.01; ranging between r = 0.26 (item 2) and r = 0.38 (item 5) for individual items). However, visual inspection revealed three clear outliers, with almost reversed scores between test and retest. When removing these outliers, test-retest correlations went up to r = 0.67 (p < 0.001) overall, and r = 0.45 (item 3) and r = 0.72 (item 4). Item-total correlations ranged between r = 0.77 and r = 0.94 at test, and between r = 0.62 and r = 0.89 at retest (Table 3). Inter-item correlations ranged between r _s  = 0.57 and r _s  = 0.77.

Good convergent validity of the scale was suggested by moderate correlations of mean trust (TiOS-SF scores) with satisfaction as assessed with the single PSQ item (M = 82.68, SD = 19.40, r _s = 0.524, p < 0.01), and with trust in the Dutch healthcare system (M = 3.90, SD = 0.83; r _s = 0.289, p < 0.05). Moreover, TiOS-SF correlated significantly with scores for recommending the oncologist (M = 4.38, SD = 1.01, r _s = 0.430, p < 0.001) and contacting the oncologist (M = 4.37, SD = 0.97, r _s = 0.310, p < 0.01).