A descriptive analysis of the contents of Care Response, an international data set of patient-reported outcomes for chiropractic patients

Advances in digital information management systems, including databases, have become important tools in understanding trends and correlations in health care by improving the collection and organisation of information such as demographics, diagnostics, therapeutic interventions, and outcomes. Routine collection of such data has now become commonplace in national health systems like the United Kingdom National Health Service (NHS) with recent imperatives focusing on the most efficient use of such data for key goals such as, improving patient outcomes, increased efficiency, and the development of effective new treatments [1].

Analysis of data collected from large cohorts of patients can have the potential to generate insights into factors identifying treatments and the characteristics of subgroups of patients who respond to certain types of care. Observational studies using large data sets have led to important public health discoveries through epidemiological analysis. For example, the Framingham study led to landmark breakthroughs in the understanding of blood pressure in disease [2].

The advent of machine learning (ML) techniques, a form of artificial intelligence (AI), is starting to change health care in profound ways, including reducing bias in diagnosis, facilitating the use of multiple sources of information to improve the understanding of pathological conditions, and improving the analysis of data sets [3]. [4] ML techniques are derived from traditional statistical methods and do not always provide clarity when trying to understand complex systems [4]. However, ML is beginning to reduce human burden in the task of diagnosis in a world of increasingly complex and interacting factors [4].

Computer technology has also made it easier to create databases that can be tailored for specialised purposes. The Care Response (CR) database has existed for ten years now and has grown steadily over time. It was designed to capture patient-reported outcome measures (PROMs) for chiropractic patients internationally. Different PROMs capture information in one or more of five different categories: quality of life related to health, functional status, symptoms, health behaviours, and patients’ perspective on their health care experience [5, 6]. PROMs can be used to monitor patient progress, guide clinical decisions, and benchmark treatment outcomes [7, 8] The use of PROMs has been associated with improved communication between clinician and patient, better symptom control, and increased patient satisfaction [9].

Specifically, the CR database was developed to help clinicians overcome some of the barriers to using PROMs regularly in day-to-day clinical practice. Electronic PROM systems are believed to reduce the administrative burden (time & cost), simplify the collection and reporting of results as well as increase the completeness of returned assessments [5, 10‐13] PROMs available on the CR system include the Bournemouth Questionnaire (BQ), Measure Your own Medical Outcome Profile (MYMOP), EQ-5D, patient global impression of change (PGIC), numerical pain scale, and Patient Reported Experience Measure (PREM) [14].

The BQ was developed to be a quick, comprehensive assessment instrument. Studies have found the BQ to be valid and reliable in back [15] and in neck [16] pain patients. MYMOP [17], is a health-related quality of life (HRQL) measure. It allows patients to nominate and score the two most important aspects of their lives that contribute most to their overall quality of life. The EuroQol 5 dimensions questionnaire (EQ-5D), is also an HRQL measure. It is commonly used for people in chronic pain [18]. Compared to some other measures, the EQ-5D has produced higher scores in healthier people, and lower scores in severely ill patients. The EQ-5D may also be more sensitive in conditions with primarily physical limitations and disability [19]. The PGIC is a 7-point ordinal Likert scale (very much improved, much improved, minimally improved, not changed, minimally worse, much worse and very much worse) [20]. The numerical pain scale (NPS) is an 11-point (0–10), commonly used pain rating tool which is very quick to administer but it is challenging to use to infer clinically important changes from baseline [21]. The PREM in the CR database is two open-ended questions: (1) Is there anything particularly good about your chiropractic care? and (2) Is there anything that could be improved?

The CR database is currently provided to clinicians as a free-to-use online system. Designed to be used in busy clinics, the advantage of the CR database is that it moves away from reliance on clinicians to provide PROMs themselves during consultations, collecting information directly from patients. After obtaining informed consent, clinics register patients by adding their name, date of birth, email address and date of first appointment either using a ’self-service’ link provided by the clinic or by clinic administrative staff. Once these fields are populated, the CR database generates a PROM questionnaire, based on the clinician’s preference, and this is provided to the patient usually via an automated email link or by the clinic on a PC or tablet device, or in paper form. Any of the available PROM instruments can be selected as a whole; questions cannot be selected from different PROMs and combined by a clinician. Multiple PROMs may be selected for use by any clinic. Subsequent PROM questionnaires are generated either at pre-set timed intervals or by the clinic manually making a request. PROM questionnaires are scored, and the clinician is presented with collated results for an individual patient or group of patients in tabulated or graphic format for longitudinal assessment after serial use of PROMs. The results of these are available immediately they are completed facilitating their use within a clinical encounter.

The CR database has been adopted in clinical practice across diverse settings and multiple countries and 218,770 patients used the CR database up to September 2022. However, up to date information on the number of clinics or countries is not available, as the system is anonymised. There are also no data on the numbers of NHS versus private patients.

Data from the CR database have been used to support multiple publications, including one of the largest UK wide cohort studies within the chiropractic profession to inform the feasibility of online PROM collection [22]. Others include the use of PROMs in clinical education, [23] studies on outcomes [24, 25] and predictors of response to chiropractic care [26, 27] There has also been a series of papers exploring the utility of the STarT Back Screening tool in professions working in the independent sector [28‐30]. Care Response data have supported 2 PhD studies, including one which resulted in the development of a theoretical model of the impact of PROMs in clinical practice [31, 32] But despite these studies there has not yet been a comprehensive assessment of the contents of the CR database.

We sought detailed participant demographics, signs and symptoms, diagnostic data as well as information on therapeutic interventions, length of treatment sessions and length of care plans. We hoped in a future study to apply ML methods to the large number of variables in the large cohort of participants to glean insight into which interventions were most effective for which conditions. This could then support clinical decision-making and policy development [33]. With information on signs and symptoms, it may be possible to discover previously undiagnosed conditions [34].

This study therefore aimed to generate a description and preliminary analysis of the entire data record, with the following objectives: to document the content of the CR database, and to explore whether the depth and breadth of information contained within the CR database would be appropriate for the application of ML analysis.

Prior to commencing data collection, data sharing agreements were drawn up between the University of Central Lancashire (UCLan) and AECC-University College (AECC-UC), and between UCLan and Clinical Transparency, Ltd, owned by Jonathan Field, which holds the CR database. Ethical approval was obtained through the UCLan Research Ethics Committee, approval #HEALTH0287. Consent to participate is obtained by Care Response and includes the option of use for research by third parties.

Two authors (KJY and JFitzgerald) first manually scanned the data to assess a broad sense of scope and completeness. The search was not date-limited, including the full 10 + years of data in the CR database. Categories of data, such as demographics, were extracted into separate tables to facilitate analysis. Without knowing what data were collected by the CR system, we were unable to determine a priori which data points we would analyse. We broadly considered that we would explore the association of demographic factors and response/completion rates. We did not know if there may be differences in responsiveness to individual items, therefore we included individual items in the analysis. Descriptive statistics were calculated for timepoints based on relevant patient characteristics available in the dataset. Based on the contents of the database, a judgement was made on the suitability of ML analysis.