Aims and objectives

This follow up cross-sectional survey of people with multimorbidity will quantify change in treatment burden over time and evaluate factors associated with treatment burden change. The survey will also facilitate evaluation of the validity of a single-item treatment burden question, modified from an item originally trialled in the baseline survey. Survey-measured, self-reported treatment burden will be compared to routine primary care metrics and the feasibility of developing a risk tool for high treatment burden based on routine data will be explored. The protocol described in this peer-reviewed article is published on protocols.io, https://dx.doi.org/10.17504/protocols.io.bw5fpg3n and is included for printing as S1 File with this article.

The objectives of the study are to:

1. Explore the degree to which treatment burden changes over time and the factors associated with such change.
2. Evaluate the validity of a single-item treatment burden screening tool versus high treatment burden at follow-up.
3. Undertake descriptive comparison of survey-measured, self-reported treatment burden with relevant metrics extracted from the primary care records of surveyed patients who provide consent for this aspect.
4. Develop and test the internal validity of a high treatment burden predictive risk tool for use in primary care/commissioning settings to identify those at risk of high treatment burden using metrics from routine clinical data of those surveyed, in comparison with survey-measured self-reported treatment burden. If secondary care data are available, then use this to further enhance the risk tool.
5. Explore whether there is any relationship between our measured change in treatment burden and the reconfiguration of services in Dorset.

Ethics

This study protocol has received Research Ethics Committee approval (reference: 21/EM/0021) and Health Research Authority (HRA) approval (IRAS project ID: 292906) on 27 January 2021.

Study sites and participants

All eight GP practices who took part in the baseline study will be approached to assist with this follow-up study. These practices were selected to represent a mixture of urban and rural, affluent and more deprived locations across Dorset. A flowchart of the methods and analysis is presented in S1 Fig.

Study participants will be patients within these practices who took part in the baseline survey and consented to receiving the follow up survey (n = 723). The original inclusion criteria of the baseline survey were those aged 55 years or over and living with at least three of the following LTCs: atrial fibrillation, coronary heart disease, heart failure, hypertension, peripheral arterial disease, stroke or transient ischaemic attack, diabetes, asthma, chronic obstructive pulmonary disease, depression, chronic kidney disease, epilepsy, osteoporosis, rheumatoid arthritis, Parkinson’s disease, multiple sclerosis, inflammatory bowel disease, coeliac disease, osteoarthritis. We defined multimorbidity as three or more chronic conditions for the following reasons: (i) to be consistent with the baseline study, (ii) the sample size calculation was based on the prevalence of a previous study which used this definition, and (iii) this has been considered to be a more valid cut-off score than the criterion of two chronic conditions for multimorbidity in elderly patients [30]. The specified diseases were selected as they are common, readily identified from GP records, and represented a range of body systems. Diseases were defined using the Quality and Outcome Framework (QoF) clinical code and Read codes [22, 23, 31, 32]. Only one person from a given household was eligible to participate [18]. Exclusion criteria were: those living in a care home, receiving palliative care, presence of a mental health diagnosis (psychosis, schizophrenia, bipolar disorder) or dementia, active cancer (recorded in the last three years), an expressed wish not to participate in research, lack of capacity to participate in the study, or any other reason at the discretion of healthcare professionals at the GP practice with sufficient knowledge of the patient which renders the patient unsuitable to receive the survey (for example if they were very frail, having cancer investigations, had a very recent bereavement It was not considered ethically appropriate or acceptable to send the survey out to patients who met the exclusion criteria as they might find it particularly distressing. Each participating site applied the same algorithm to electronically search and identify potential participants. Manual screening was performed by a health professional to exclude patients that met the exclusion criteria. A maximum of 250 patients (randomly selected from all potentially eligible participants) from each practice were invited to participate in the baseline survey.

Participant recruitment and invitation

Participating practices will be asked to identify the patients who were participants in the baseline survey and consented to receiving the follow-up survey, using a list of unique study IDs. Lists of these potentially eligible patients will be manually screened by GPs (or another suitable healthcare professional with sufficient patient knowledge from the GP practice) for exclusion criteria (as some participants may now meet the exclusion criteria). Practices will be encouraged to only exclude people if they feel it is absolutely necessary, while acknowledging that in some circumstances, sending the follow-up survey may be inappropriate.

Survey items

For this follow-up survey, changes from the baseline survey design are minimal in order to provide maximally comparable follow-up data from that acquired at baseline. Survey questions and response options of the baseline survey are available elsewhere [18].

Data extraction from primary care records

Data from primary care records will be extracted for respondents who consent to data from their GP record being shared with the research team. Extracted data will include: demographic details (current age, sex, full postcode) as recorded in primary care records, frailty score (electronic Frailty Index and/or Rockwood score where available), main diagnosed health conditions as defined by Read codes, medications on repeat prescription (including details of dose, frequency, and route of administration), recent healthcare usage data (including number of GP practice appointments by type -surgery appointment, telephone consultation, home visit- and role (GP, practice nurse, or other team member), number of outpatient appointments/referrals to secondary care, accident and emergency attendance (if available) and hospital admission data (a count of discharge summaries), as available in the primary care record. We will discuss with colleagues in the CCG about the potential for enhancing primary care data with data from secondary care. These variables have been selected as they may be expected to be associated with high treatment burden.

Sample size

The pool of participants is limited to the 723 individuals who consented to receiving the follow-up survey. The sample size of the baseline survey was determined using nQuery based on prevalence of high treatment burden (26.6%) from the validation study of the MTBQ together with estimates of likely response rate of approximately 50% (based on the literature) and distribution of the MTBQ score [18]. It was estimated that a minimal sample size of 150 participants would be needed to detect the minimum possible change in treatment burden score (1 point in the 40-point MTBQ raw score) over time. A follow-up sample of 361 participants is anticipated, assuming a likely follow-up survey response rate response rate of 50%.

Statistical analyses

Descriptive statistics will be used to compare the numbers and proportions of participants reporting ‘none’, ‘low’, ‘medium’ and ‘high’ levels of treatment burden based on the global MTBQ score (score of 0, <10, ≥10, ≥22, respectively). We will consider those who report a ≥1 point change in the 40-point MTBQ raw score as having experienced a change in treatment burden, though we recognise this change may not be clinically meaningful. Change in global score and change in these categories between baseline and follow-up will be compared. Characteristics (such as age, sex, education status, and comorbidities) of participants who experience change in treatment burden, those who experience an increase in treatment burden, those who experience a decrease in treatment burden and those who don’t will be described. We will also describe and compare characteristics of those who have experienced an increase in treatment burden to those who have not and those who have experienced a decrease in treatment burden to those who have not. Normally distributed variables will be compared using t-test and categorical variables will be compared using chi-squared tests. Associations with change in categories of treatment burden (i.e., overall change vs no change; decrease vs no change or increase; increase vs no change or decrease) will be examined using a linear mixed model that additionally adjusts for potential confounders. We will also examine effect modification using interaction terms. We will also explore geographical access aspects of treatment burden, considering issues such as distance and travel time to GP and hospital services and distance /time from pharmacies (and/or whether individuals are registered with dispensing practices). Further consideration will be given to availability of public transportation for any of these services. Distance to services for this population may be contrasted with access to treatment for similar age/sex populations in other areas of Dorset to assess whether there is significantly better/worse service provision for the study population. Regressions will adjust for time (year) to account for (and explore impact of) changes to services in Dorset.

The single-item screening tool (at follow-up) will be evaluated relative to high treatment burden as indicated by the MTBQ (at follow-up) using sensitivity, specificity, positive and negative predictive values, and positive and negative likelihood ratios. Area under the receiver operator characteristic curve will also be assessed.

To develop a population-level high treatment burden risk prediction tool, variables available in routine data (such as number of prescribed medications, number of appointments, number of hospital admissions (if available), number of LTCs) will be included in a first stage of a binary logistic regression model predicting risk of high treatment burden based on the MTBQ measure (a score ≥22 indicating high treatment burden versus a score <22 indicating non-high treatment burden). After assessing extent of missing data, a forward-stepwise, likelihood-ratio model building approach will be employed, with only those variables having a statistically significant (p<0.05) effect on the final model equation remaining at the terminal step [34]. A binary logistic regression model containing only these variables will be fitted and weighted scores based on the beta coefficients derived. Model performance will be determined using the area under the receiver operator characteristic curve (to assess discriminant capacity of the model) and calibration plots (to assess calibration) [34].

Data analyses will be led by a public health researcher (HH) with input from a senior medical statistician (SH). All authors (comprising senior academics, academic GPs, Director of Public Health, Chair of the Clinical Commissioning Group, specialist registrars in public health and elderly care, and public contributors) will be involved in interpretation of analyses and results.

Mapping changes in primary and care services

Working with leaders of the Clinical Commissioning Group we will identify the key changes in primary and secondary care delivery that have occurred in the intervening period including structural issues (such as infrastructure change and location of services and workforce), process issues (such as change in care pathways, virtual vs. in person clinical encounters) and outputs (such as changes in numbers of patients seen in different settings).

Patient and public involvement

Two public contributors, both living with multiple LTCs, have been involved in the design of the study and are co-investigators. They have reviewed the lay summary and provided valuable insights into the challenges of living with multiple conditions, including the impact of treatment workload on social care. Both will continue engagement with the project throughout its duration and will contribute to outputs and dissemination.

Dissemination

Study findings will be published in peer-reviewed academic journals and presented at relevant national and international conferences. We will engage with relevant stakeholders to disseminate findings throughout Dorset. Opportunities to implement a high treatment burden risk tool in practice will be explored.

Strengths and limitations

Strengths of the study include a reasonable likelihood of a high response rate given that participants are those who responded to a similar baseline survey, the inclusion of participants from diverse socioeconomic backgrounds and the use of a validated measure of treatment burden in the survey. However, there are important limitations that should be considered. The first is the potential impact of the COVID-19 pandemic on the follow-up survey. In view of the pandemic, the timing of distribution of the follow-up survey has been delayed until late summer 2021 to minimise any potential impact as well as to allow GP practices more time to engage with the study. Despite this, the pandemic has the potential to influence the survey in a number of ways. Firstly, participants may themselves have had COVID, which might have increased their recent experience of ill-health and treatment burden. Secondly, changes to healthcare delivery as a result of the pandemic may mean that most participants will have had reduced or altered healthcare appointments and use of other healthcare resources, which may influence their responses. This will potentially make comparison to baseline survey responses challenging, particularly for the survey section about healthcare usage but also potentially to the treatment burden and quality of life questions. However, the effect on responses to the MTBQ is likely to be small due to the nature of the MTBQ questions, which ask about current rather than past experiences. Other limitations include the inability to infer causality from any observed associations. It is also possible the screening of consented patients for participation could lead to a sample that is skewed towards the lower end of the treatment burden scale. Finally, a possible limitation is that metrics available in routine data may not sufficiently reflect experiences of treatment burden. This may be particularly true given high treatment burden was associated with markers of capacity (health literacy, financial resource) [17], which are not routinely collected. However, this will be an important part of the exploration of the feasibility of developing a high treatment burden risk tool.