Study design and participants
CREW is a multicentre, prospective cohort study of newly diagnosed patients with CRC, treated with curative intent in 29 cancer centres across the UK. Patients were eligible if they (a) had a diagnosis of non-metastatic CRC (Dukes’ A–C), (b) were awaiting primary surgery with curative intent (patients who had been identified as eligible and admitted for emergency surgery were also included), (c) ≥ 18 years old, and (d) had the ability to complete questionnaires. Recruitment took place between November 2010 and March 2012; the aim was to include every eligible patient diagnosed at each cancer centre during the centre’s recruitment period. Further details relating to eligibility, recruitment strategy, and sample size are provided elsewhere .
Written consent was obtained and baseline questionnaires completed prior to surgery whenever possible. Socio-demographic information including gender, age, ethnicity, employment, and domestic status was also collected at consent. Clinical details including tumour site (colon, rectum), Dukes’ stage (A, B, C1, and C2), and treatment (neoadjuvant, adjuvant, and stoma formation) were extracted from medical records at 6 months and verified at 24 months post-surgery. Self-reported comorbidities were recorded from 3 months post-surgery.
Patient self-report questionnaires also measured a broad range of issues including supportive care needs, life events (e.g. spousal/relational problems, bereavement, financial problems), health status, HRQoL, wellbeing, social support, and self-efficacy. Follow-up mailed questionnaires were completed at 3, 9, 15, and 24 months post-surgery (longer term data collection continued for up to 5 years).
The study was approved by the UK National Health Service National Research Ethics Service (REC reference number: 10/H0605/31).
Full details of the repeated measures used in the CREW study are provided elsewhere .
The SCNS SF-34  was used to assess perceived needs and whether they had been met at 15- and 24-month follow-up points. The scale includes 34 items covering five domains of need: physical/daily living, psychological, sexuality, patient care and support, and health system and information. Each need is rated on a 5-point Likert-type scale with responses 1 (“Not applicable” or “No need”) and 2 (“Need satisfied”) indicating no outstanding need while 3, 4, and 5 measure degree of unmet need ranging from “Low” (little need for additional help) to “High” (strong need for additional help).
The incidence of significant life events in the past 6 months was recorded from 3-month follow-up using a modified version of the list of threatening life experiences , with death of a spouse and child presented as separate life events and death of a pet and moving house added.
Health status, HRQoL, and wellbeing
The EuroQol 3-level version (EQ-5D-3L)  measures generic health status across five domains: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression each scored as none/some/severe problems.
The Quality of Life of Adult Cancer Survivors (QLACS) Scale Part 1  measures 28 QoL issues across eight generic domains which are not necessarily attributable to cancer: negative feelings, positive feelings, cognitive problems, pain, sexual interest and sexual function (merged together), energy/fatigue, and social avoidance.
The European Organisation for Research and Treatment of Cancer (EORTC) 30-item Core measure (QLQ-C30)  and the CRC module (QLQ-CR29)  were included from 3 months onwards. The QLQ-C30 includes five functional scales, three symptom scales, six single items, and a Global health status/QoL scale. The QLQ-CR29 supplements the QLQ-C30 with 29 colorectal cancer-specific items, from which are derived four functional subscales (body image, anxiety, weight, and sexual interest) and 19 single item scales. A linear transformation is applied to produce subscale scores with a possible range from 0 to 100, with high scores indicating better functioning on the functional and Global health/QoL scales but poorer symptoms on the symptom scales/items.
The Personal Wellbeing Index–Adult (PWI-A)  includes eight items of satisfaction corresponding to standard of living, health, achieving in life, relationships, safety, community connectedness, future security, and spirituality/religion.
Anxiety and depression were assessed using the State-Trait Anxiety Inventory (STAI)  and the Centre for Epidemiologic Studies Depression Scale (CES-D) .
Positive and negative mood states were measured using 10 questions (5 questions for each mood state) from the Positive and Negative Affect Scale (PANAS) .
The MOS Social Support Survey (MOS-SSS)  provides an indication of how often different types of support are perceived to be available including emotional/informational, tangible, affectionate, and positive social interaction.
The Self-efficacy for Managing Chronic Disease Scale (Lorig)  includes six items measuring confidence in different areas.
Univariate analysis of baseline characteristics of participants was conducted using a chi-squared test to compare those who completed at least one SCNS domain at 15 months with those who did not. A multivariable logistic regression model was additionally constructed to assess the differences once all significant characteristics were mutually adjusted for (p < 0.05).
Independent variables were divided into five thematic blocks: (1) socio-demographic characteristics (gender, age, employment, and domestic status), (2) clinical details (tumour site, staging, treatment, stoma, co-morbidities), (3) psychosocial measures (EQ-5D, QLACS-GSS, PWI-A, CES-D, STAI, PANAS, MOS, Lorig), (4) cancer HRQoL (QLQ-C30), and (5) CRC-specific HRQoL (QLQ-CR29). Due to a skewed distribution in most of the psychosocial measures and EORTC subscales, all scores, with the exception of QLACS-GSS and the QLQ-C30 Global health/QoL score, were converted into categorical covariates. See Supplementary Material 1: Table 1.2, footnote 1 for details.
Descriptive measures were used to examine the prevalence of total unmet supportive care needs. Consistent with previous research on unmet needs , the number of participants with no, few (1–4), and many (5 or above) moderate or severe unmet needs was calculated. In terms of missing data, patients were identified as having 5 or more unmet needs even if some of the composite questions were missing (i.e. at least 5 were rated as severe or moderate), otherwise the data were excluded from the analysis. To investigate change in perceived needs over time, a chi-squared test was used to test for differences (marginal homogeneity) in the distribution of both overall and SCNS domain unmet needs between the two time points of 15 and 24 months. In addition, the ten most common moderate/severe unmet needs at 15 and 24 months post-surgery were ranked and compared.
To predict which individuals are more likely to have unmet needs at the end of treatment (15 months post-surgery), logistic regression models were used to assess the baseline predictors of unmet need for each SCNS domain. In line with previous research using the SCNS [16, 21], each SCNS domain of unmet needs was dichotomised into no or low unmet need (score = 1 to 3) and moderate or severe unmet need (score = 4 or 5). Independent variables were taken at the earliest time point available which was at baseline except for the comorbidity status and the EORTC measures (QLQ-C30 and QLQ-CR29), which were first completed at 3 months. Therefore, we restricted our sample to those respondents who participated at three time points: baseline, 3 months, and 15 months. For each SCNS domain, the final regression models were adjusted for age as an important demographic effect-modifier.
To investigate the associations between the SCNS domains of unmet need following treatment (15 months) and HRQoL outcome (also at 15 months), an ordinary least squares (linear) regression model was produced. The QLQ-C30 Global health/QoL scale score at 15 months was preferred as the (continuous) outcome measure since there was no evidence against the assumption that the resulting residuals for the linear regression model were normally distributed. All five SCNS domains (physical/daily living, psychological, sexuality, patient care and support, and health system and information) were the covariates of interest at 15 months and entered into the model together. The model was adjusted for the socio-demographic and clinical characteristics (Blocks 1–2, including disease recurrence and negative life events). Blocks 3–5 were not included because of the hypothesised interaction between these variables and supportive care needs. Additional regression models were produced to explore the association between each individual SCNS domain and HRQoL outcome at 15 months, controlling for the significant confounders.
Random clustering effects of cancer centre were assessed by fitting centre separately as a fixed variable to each regression model and as a random effect. Both effects were negligible for all models; therefore, centre was not included in subsequent models.
For all regression models, a forward stepwise selection approach was used, which was applied in two steps: first, separately to each thematic block of independent variables; second, to the final set of those covariates which remained significant in each block.
The significance level was fixed at 5% and analyses were carried out using Stata Corp. StataSE 14 and IBM SPSS Statistics 24.