Baseline data were available for all the secondary outcomes.

Analysis of secondary outcomes

Patient-level multiple regression analysis of complete cases.

Dependent variable: secondary outcome.

Independent variables: trial arms; baseline value of secondary outcome; pre-specified covariates.

Level of significance: 0.05.

Adjustment of significance and confidence levels due to multiplicity of outcomes The analysis of secondary outcomes is regarded as exploratory and therefore not subjected to adjustment for multiple testing.

Alternative tests if distribution assumptions are violated Regression with bootstrapping.

Sensitivity analysis See relevant section below.

5. Sensitivity analyses

All the above analyses will be subjected to three sensitivity analyses.

The first (and main) sensitivity analysis will repeat the primary analyses using multiple imputation to include cases with missing baseline or follow-up data (see below).

The second sensitivity analysis will assess the robustness of the main analysis results to changes in the regression model. This analysis will remove the pre-specified covariates from the model (but still include the outcome at baseline).

A (provisional) third sensitivity analysis will examine the results after excluding patients with very short time intervals between the date receiving the first intervention phone call and return of the 20-month questionnaires. This is provisional because we will precede this by an analysis to examine the variation in times and only proceed with the sensitivity analysis if there are substantial numbers with intervals of <6 months.

For the main sensitivity analysis, missing data values for variables atbaselineandfollow-upwill be substituted using chained-equation multiple imputation (MI) procedures. We will apply multiple imputation to baseline and 20-month variables with missing values by the chained equations approach using scores on all primary and secondary outcome measures (at baseline and follow-up). We will use 20 multiple imputation sets, as this will provide appropriate stability of results. Cases for whom imputation of baseline values is not possible (e.g. where the entire baseline questionnaire is missing) will be excluded.

7. Distributional tests

We will examine the distributional properties of each outcome variable. Variables for which skewness or kurtosis>1.0 will be analysed using a bootstrap method. We will not do tests for normality because the large sample size makes these likely to be significant even for small deviations from normality.

8. Bootstrapping

Bootstrapping ofp-values and CIs will be applied for outcome variables with skew>1.0 or kurtosis>1.0. In these cases the bootstrapped estimate of standard error will be used. Prior to any bootstrapping a set of pseudorandom numbers will be generated (depending upon how many outcomes have skew or kurtosis >1.0) using random.org to act as seeds for each bootstrap analysis.

9. Choice of covariates

The covariates to be included in all primary and secondary analyses will be selected in the below manner.

First, the baseline values of the outcome that is the focus of each analysis will be included as co-variates. Second, a set of pre-specified covariates will automatically be included. The categorisation of variables (e.g. age, number of long-term conditions) is based upon examination of the distributions of these variables at baseline. These variables will be included in all primary and secondary analyses to reduce bias resulting from missing data.

TABLE 46 Primary analysis 1 (outcome patient activation)

Pre-specified covariates Description

Sex Question A1GENDER

Coded as Male or Female

Age Question A2AGE

Recoded as Agecat2 65–69; 70–79; 80–98

General Practice ID GP practice ID number

Health literacy (baseline) A single 1–5-item health literacy measure Baseline social support (ESSI)

(from baseline)

The ESSI is a 7-item scale. A total score is calculated by summing all individual items; a higher score indicates greater social support

Baseline patient activation PAM total continuous score

Baseline depression MHI-5

Coded as probable depression>60; no depression<60

TABLE 47 Primary analysis 2 (outcome WHOQOL physical domain)

Pre-specified covariates Description

Sex Question A1GENDER

Coded as Male or Female

Age Question A2AGE

Recoded as Agecat2 65–69; 70–79; 80–98

General Practice ID GP practice ID number

Health literacy (baseline) A single 1–5-item health literacy measure

Baseline social support (ESSI) (from baseline) The ESSI is a 7-item scale. A total score is calculated by summing all individual items; a higher score indicates greater social support

Baseline patient activation PAM total continuous score

Baseline depression MHI-5

Coded as probable depression>60; no depression<60

Baseline Quality of life WHOQOL physical health domain

TABLE 48 Secondary analysis 1 (depression)

Pre-specified covariates Description

Sex Question A1GENDER

Coded as Male or Female

Age Question A2AGE

Recoded as Agecat2 65–69; 70–79; 80–98

General Practice ID GP practice ID number

Health literacy (baseline) A single 1–5-item health literacy measure

Baseline social support (ESSI) (from baseline) The ESSI is a 7-item scale. A total score is calculated by summing all individual items; a higher score indicates greater social support

Baseline patient activation PAM total continuous score

Baseline depression MHI-5

Coded as probable depression>60; no depression<60

Baseline Quality of life WHOQOL physical health domain

Pre-specified covariates Description

Sex Question A1GENDER

Coded as Male or Female

Age Question A2AGE

Recoded as Agecat2 65–69; 70–79; 80–98

General Practice ID GP practice ID number

Health literacy (baseline) A single 1–5-item health literacy measure

Baseline social support (ESSI) (from baseline) The ESSI is a 7-item scale. A total score is calculated by summing all individual items; a higher score indicates greater social support

Baseline self-care activities SDSCA total

Baseline depression MHI-5

Coded as probable depression>60; no depression<60

Baseline Quality of life WHOQOL physical health domain

TABLE 50 Primary outcomes

Primary outcome

(all at 20 months) Description

Patient Activation Measure

A standardised spreadsheet in excel is used to score the PAM. Overall score (0–100) on the PAM scale where higher scores indicate high patient activation. A total score will be generated where participants answer at least 10 out of the 13 PAM questions. The continuous scores are categorised into four levels for descriptive purposes using the standardised spreadsheet. Higher scores indicate higher patient activation

Quality of life–

physical health domain

Physical health domain of the World Health Organization Quality of Life instrument (WHOQOL-100). WHOQOL is a 26-item measure which includes two items on general overall QoL and health, plus 24 which are scored in four QoL domains: physical, psychological, social relationships and environmental QoL. Facets are scored from 1 to 5, and higher scores indicate better QoL. Domain scores are transformed onto a scale from 0 to 100

TABLE 51 Secondary outcomes

Secondary outcome Description

Depression measure The MHI-5 is a five-item scale which measures general mental health, including depression, anxiety, behavioural-emotional control and general positive affect. The standardised score ranges from 0 to 100; scores below 60 indicate probable depression

Baseline SDSCA A 7-item measure which assesses the number of days per week respondents engage in healthy and unhealthy behaviours (i.e. eating fruit and vegetable, eating red meat, undertaking exercise, drinking alcohol, and smoking)

In document Improving care for older people with long-term conditions and social care needs in Salford : the CLASSIC mixed-methods study, including RCT (Page 175-179)