The use of sequence analysis to study primary care pathways: an exploratory study of people at high risk of lung cancer in England

Abstract

Research background, gaps, and aim: Lung cancer (LC) is a research priority in the UK, due to its high incidence and mortality, and poor survival. Current population-based studies are focused on investigating ‘route to diagnosis’, factors associated with late diagnosis and poor survival, and the implications of different intervals (e.g. primary care interval, diagnostic interval, treatment interval) in the cancer care pathway. However, the longitudinal sequence of interdependent patient-GP events over time (patient’s help-seeking behaviours and general practitioner (GP) management) preceding cancer diagnosis is a research area less investigated. Therefore, this PhD study proposes a new perspective of studying primary care sequences for early diagnosis research, using a novel statistical method – sequence analysis (SA), to identify meaningful typologies in a less investigated population – patients at high risk but not yet diagnosed with LC. Methodology: A systematic scoping review was conducted to understand how SA has been applied to study disease trajectories and care pathways in health services research, to learn the lessons from published studies and inform the application of SA in the main study.
- Study design, setting, and participants: 899 community patients at high risk of developing LC (based on patient's smoking history) but not yet diagnosed with LC from eight general practices in the South coast of England consented to participate in this study. Their primary care records from June 2010 to October 2012 (29 months) were reviewed. Information was extracted from GP notes in free text and transcribed manually.
- Research process: Two study phases, methodological exploration and empirical analysis, were involved to address three research objectives: how to construct primary care sequences from discrete health events, how to use different features of SA to obtain meaningful cluster patterns (the outcome of SA), and how patients’ sociodemographic and clinical characteristics can help explain the variation in the cluster patterns and help-seeking behaviours.
- The primary outcome, covariates, and statistical methods: SA and cluster analysis were used to obtain the primary outcome – typology of clusters. Descriptive statistics were used to characterise patient profiles for each cluster, followed by traditional statistical tests (ANOVA, chi-square test, Kruskal-Wallis test) to compare patients’ sociodemographic and clinical features among clusters. Generalised linear models were used to explore the association between patient characteristics and clusters, and to quantify the relative risk ratios. Key findings: The study sample was classified into seven clusters. The subgroups of patients who presented with potential LC symptoms were categorised into four clusters with different GP management: GP ordered tests or prescribed medications for transient symptoms (Cluster 1, n=133/899, 14.8%); GP ordered chest X-ray or referred patients to specialists (Cluster 2, n=65, 7.2%); GP offered health advice to patients (Cluster 3, n=60, 6.7%); and patients presented symptoms multiple times and received repeated prescriptions from GP (Cluster 4, n=37, 4.1%). Cluster 5 was patients without potential LC symptoms but presented in general practice with cardiorespiratory comorbidities and/or other alarm symptoms (n=326, 36.3%). Patients in Cluster 6 only had minor care needs (n=237, 26.4%). Patients in Cluster 7 did not visit GP at all during the whole study period (n=41, 4.6%). For patients who had ≥2 visits with potential LC symptoms, the median interval was 61.5 days, interquartile range [16, 217] days. Age and the number of comorbidities were the two significant variables in different models. Variation of patient management among practices (practice effect) was observed. Conclusions, clinical relevance, and implications: This PhD thesis has made an original contribution by establishing the feasibility and analytical framework of using SA to study complex primary care sequences in the field of early diagnosis research. This study demonstrates the potential of applying SA in a larger scale study with a more representative population, to investigate the complex and heterogeneous primary care sequences leading to LC diagnosis, which has clinical implications for patient care and management, promote early cancer diagnosis from primary care, and eventually improve LC survival in the UK.

More information

Identifiers

ORCID for Weiqi Liao: orcid.org/0000-0002-8605-3749

ORCID for Lucy Brindle: orcid.org/0000-0002-8933-3754

Catalogue record

Export record