Based on the suggested operationalization of co-care as a distributed system of activities, we set out to develop a scale to measure this and to evaluate the scale’s psychometric properties.
DoCCA scale development
Initially, several instruments in the patient and user experience literature were scrutinized to identify suitable measures for the operationalization of co-care as a distributed system of activities (e.g., [13, 14, 17, 38,39,40,41,42,43]). However, as discussed, these did not sufficiently reflect the system perspective of chronic care management and its distributed nature. Thus, we turned to instruments from the work psychology domain, primarily the Copenhagen Psychosocial Questionnaire (COPSOQ) and Bern Illegitimate Tasks Scale (BITS). COPSOQ is a validated instrument to measure psychological work conditions related to work tasks, the organization of work, and interpersonal relations, and has been widely used in both practical settings and in research [44]. BITS measures how individuals perceive the tasks they are expected to perform [45]. It consists of two sub-scales that measure the degree to which tasks are perceived as unnecessary or unreasonable.
Item pool generation and selection
An initial item pool was generated from the COPSOQ and BITS instruments through an iterative process. First, we examined all subscales and items for their relevance for the co-care construct (Fig. 1), testing whether they made sense when rephrased to suit the context of chronic care management rather than a work setting (e.g., references to managers, colleagues, and work tasks were rephrased to references relevant to chronic care management viewed from a system perspective). Forty-three candidate items were identified in the first iteration. The number of items were reduced iteratively through discussions among the authors about individual items’ relevance, applicability, acceptability, and appropriateness and the combined coverage of the co-care concept.
This process yielded 14 items, 12 of which were related to distribution of work tasks. They came from three subscales: Cognitive and emotional demands, measuring needs to make difficult decisions, remembering or monitoring things, and exposure to emotionally strenuous situations; Role clarity, measuring how responsibility for activities is distributed, perceived clarity of responsibilities and expectations, and whether the distribution is perceived as just; Unnecessary tasks (from BITS), measuring tasks that do not need to be performed. To reflect the factor structure from the original scales, we thus propose that distribution of activities consists of three sub-factors (demands, role clarity, and unnecessary tasks). We found two items of relevance for needs support, one from Social support, assessing whether one receives the support needed, and one from Predictability, namely if one receives information needed to perform tasks.
Generation of new items and refinement
With only two items reflecting needs support and none of relevance for goal orientation, we returned to the patient satisfaction survey used in Swedish healthcare [46] and adapted two items, one assessing participation in decision making and one for information clarity. Still not finding any items reflecting the perspective shift evident in the conceptualization of the goal orientation of the system, we then created four new items in an iterative process where all authors were engaged.
Feedback from nine individuals (representing individuals with chronic conditions, eHealth-experts, healthcare professionals, and researchers) was elicited at different time points in the development process. We also performed cognitive interviews with three individuals with different chronic conditions (hypertension, chronic pain, diabetes) to assess item comprehension and the relevance of the final version, leading to minor language edits.
The DoCCA Scale
The final DoCCA scale consists of 20 items. For all items, a five-point Likert response scale was used, consistent with COPSOQ, with the following anchors: 1 = To a very low degree; 2 = To a low degree; 3 = Partially; 4 = To a high degree; and 5 = To a very high degree. The exceptions were items derived from COPSOQ´s cognitive and emotional demands scales, which included the original response scale: 1 = Never/Almost never; 2 = Seldom; 3 = Sometimes; 4 = Often; and 5 = Always. Note that for items concerning demands and unnecessary tasks, a low value is positive, and for the others, a high value is positive. The questions were formulated to capture the present (e.g., “Do you feel that …”) in line with suggestions for capturing experience data, and to minimize the risk of recollection bias [47].
Psychometric testing
Setting
The DoCCA scale was tested in a Swedish primary care setting in which a pilot test of an eHealth service was conducted that entailed shifting tasks and activities from primary care providers to patients, such as for blood pressure measurements. The eHealth service consisted of monitoring devices and a smartphone application. Measurements, as well as trends and alerts, were automatically communicated to the individuals and their primary care staff. Asynchronous communication through chat was also supported.
Recruitment
Participants were recruited through the primary care organization. The eHealth service targeted adult patients (≥ 18 years of age) diagnosed with hypertension, chronic heart failure, or mental health conditions, including reaction to severe stress and adjustment disorders, insomnia, anxiety disorders, and depressive disorders. Individuals with one or more of these chronic conditions were eligible. Further inclusion criteria for participating in the pilot test were having a smartphone and email account and being able to communicate in Swedish, since this was the language used in the smartphone application.
Primary care staff identified patients that met the diagnostic inclusion criteria and called them by phone to inform them of the project. Interested and eligible participants were invited to a group enrollment session at the primary care clinic in September and October 2018. During the session, they were informed about the research project by a member of the research team and were invited to participate. Informed consent was obtained. The project followed the guidelines of the Helsinki Declaration and was approved by the Regional Ethical Review Board of Stockholm (reg nr. 2018/625-31/5 and 2018/1717-32).
Participants and data collection
Data were collected at two time points using a web-based questionnaire after enrollment in the eHealth project and seven months later. Thus, data were collected both when participants had recently been introduced to the eHealth service, hence assessing a co-care system during more traditional chronic care, and after they had personally used the eHealth service, allowing for validation of co-care in the same population with and without significant eHealth components. The questionnaire contained several measures (list available on demand). Here, we report those used in the psychometric testing of the DoCCA scale.
The questionnaire was distributed to 308 recipients. Reminders were sent out one and two weeks after the initial mailing. The response rate was 174 (56 %). The second questionnaire was distributed to the same 308 recipients after seven months. A total of 134 responded (after two reminders), yielding a response rate of 44 %. One-hundred and thirteen respondents (37 %) provided complete answers to both questionnaires. Of these, 81 (72 %) reported using the eHealth service to manage their hypertension, 21 (19 %) for mental conditions, 20 (18 %) for other (unspecified) chronic conditions, and nine (8 %) for heart failure. Four respondents (4 %) did not know what they used the eHealth service for. Twenty respondents (18 %) reported that they suffered from comorbid chronic conditions.
Statistical analysis strategy
The analysis included four parts. First, the psychometric properties of the DoCCA scale were investigated by evaluating construct validity with Confirmatory Factor Analysis (CFA). The CFA tested whether the items loaded on the intended dimension. Each set of items was allowed to load only on its corresponding latent variable. No correlation errors either within or across sets of items were allowed. However, in line with the theoretical assumptions of the measurement instrument, non-zero correlations between the factors were allowed. The proposed models assumed three factors: activities, needs support, and goal orientation, with the first additionally divided into sub-factors of demands, unnecessary tasks, and role clarity. This second-order (SO) structure was compared to three alternative solutions: (1) a one-factor (1 F) model with all items loading onto one factor, (2) a three-factor (3 F) model where demands, unnecessary tasks, and role clarity formed one factor (i.e., no sub-factors within activities) and needs support and goal orientation formed two separate first-order factors, and (3) a five-factor (5 F) solution with demands, unnecessary tasks, role clarity, needs support, and goal orientation as independent factors.
In line with the multifaceted approach to assessment of model fit, we considered the following fit indices: Comparative Fit Index (CFI) [48], Tucker and Lewis Index (TLI) [49], Root Mean Square Error of Approximation (RMSEA) [50], along with 90 % confidence interval limits, and (Standardized) Root Mean Square Residual ([S]RMR) [48]. We used the following values as thresholds recommended in the literature: TLI and CFI > .90 [51], RMSEA < .08 [50], and (S)RMR < .08. We followed the steps described for T1 data. Time 2 data served as a cross-validation, where we tested the goodness-of-fit indices of the model chosen in step 1. CFA analyses were performed in Mplus [52].
Second, the scale’s validity was examined. We computed Pearson’s moment correlations between the identified factors. To test concurrent validity, we computed cross-sectional correlations between the subscales of our instrument and a previously-validated measure of patients’ experiences managing a chronic condition, the six-item Self-Efficacy in Self-Care (SESSC) scale [53]. The SESSC was developed to measure self-efficacy in the context of minor illness. It includes questions such as: “How certain are you that you can:”, e.g., “affect your symptoms?”, “regulate your activities so as to be active without aggravating your symptoms?”. Responses were made on a four-point scale (1 = very uncertain and 4 = very certain). Self-efficacy and co-care as a distributed system of activities are both constructs, that intend to reflect people’s experiences of managing a chronic condition, but in different ways. While DoCCA reflects the individual’s experience of the co-care system, SESSC reflect the individual’s confidence in his or her own ability to manage the condition. Thus, we do not expect perfect correlation between these measures, but, overall, that individuals with high self-efficacy scores would also have positive experiences of the distribution of co-care activities, and vice versa.
Third, as a test of predictive validity, we tested correlations between the dimensions of the DoCCA scale measured at T1 with satisfaction with healthcare measured at T2. A one-item satisfaction question was used, based on the Swedish national patient survey: “What is your overall rating of the care you have received at the [name of primary care center] during the past 6 months?”. Responses were made on a five-point scale (1 = bad, 2 = reasonable, 3 = good, 4 = very good, 5 = excellent). The expectation was that a positive experience in DoCCA (i.e., low demands, low unnecessary tasks, high role clarity, high needs support, high goal orientation) at T1 will predict better satisfaction with healthcare services at T2. Kendall’s tau-b (τb) correlation coefficient was computed since satisfaction was measured on an ordinal scale.
Fourth, reliability analyses of the subscales were conducted by assessing the internal consistency of the subscales by calculating Cronbach’s alpha coefficients. Test-retest reliability was analyzed by computing Pearson’s moment correlations for each subscale based on the sample of 113 respondents for whom both T1 and T2 data were available.
In all analyses, missing data were deleted list-wise.