Asma Chaabouni

Introduction: Symptom diagnosis is defined by the Word Organization of Family Doctors (WONCA) international classification committee as ‘using a symptom or complaint as the best medical label’ for the presented health problem. Symptom diagnoses are highly relevant to primary care. Although symptom diagnoses constitute an important part of the workload of General Practitioners (GPs), managing patients with symptom diagnoses is still a challenge for GPs, particularly those with persistent symptoms. Even though World Organization of Family Physicians (WONCA) highlights that symptom research as an evidence gap in the literature, symptom diagnoses is still under-researched in general practice.

Aims: We aimed to provide evidence-based knowledge on morbidity rates, the burden experienced by patients with persistent symptoms, the course and GPs management of symptom diagnoses. In addition, we aimed to identify patients at a high risk of developing persistent symptoms and explore the potential of Artificial Intelligence (AI) in determining diagnostic prediction models in primary care.

Methods: We analyzed data from primary care Electronic Health Records (EHRs) and surveys, integrating perspectives from both daily clinical practice and epidemiological research. Based on data-driven and experts opinions, we defined persistent symptoms as symptoms that lasted for more than a year from the first until the last contact with the GP. We performed a systematic review to examine the potential of Artificial Intelligence (AI) in developing diagnostic prediction models utilizing EHRs.

Main findings:

1. Prevalence of (persistent) symptom diagnoses
In Chapter 2, we conducted a retrospective cohort study to examine the ecology of symptom diagnoses in primary care using data from the Family Medicine Network (FaMe-Net) database. We found that symptom diagnoses were highly prevalent as 57.9% of patients visited their General Practitioners (GPs) for at least one symptom diagnosis in one year. The incidence rate of symptom diagnoses in 2018 was 767 episodes per 1000 patient-years. Of these, approximately one in six had a symptom diagnosis that persisted for more than a year. Symptoms that persisted for more than a year were more common among women, older adults, and patients with multiple comorbidities. In Chapter 3, we extended this analysis to children, utilizing the FaMe-Net database. We found that symptom diagnoses were also prevalent as almost half of the children having at least one GP visit for a symptom diagnosis within a year. However, symptom diagnoses lasting over a year were less frequent in children than in the general primary care population, affecting only one in 20 children.

2. Course and management strategies for symptom diagnoses
In Chapter 4, in a retrospective cohort study, we examined the course of symptom diagnoses and found that most episodes starting with a symptom diagnosis were transient (85.8%), with fewer than 10% evolving into somatic conditions and only 4.3% persisting for over a year. In terms of management, GPs used more intensive strategies in the first year for symptoms that evolved into somatic conditions compared to those that remained persistent. Symptoms diagnoses that persisted for more than a year were mainly managed within primary care. In Chapter 5, we conducted a retrospective cohort study focusing on the course and GPs’ management strategies of psychological symptom diagnoses. We found that 77.5% of episodes were transient, while 12.8% persisted for more than a year, and only 7.8% developed into psychiatric conditions. Overall, GPs used more management strategies for episodes that evolved into psychiatric conditions compared to those that persisted as psychological symptoms, during the first year. However, for persistent psychological symptoms, more referrals were indicated compared to episodes that evolved into psychiatric conditions. Given our findings in Chapters 4 and 5 that most symptoms are transient, we stress the importance of reassurance of patients presenting with symptom diagnoses and psychological symptom diagnoses when discussing the prognosis.

3. The burden of persistent symptom diagnoses
In Chapter 6, we explored in a cross-sectional study the level of burden in patients having symptom diagnoses for more than a year. We found that around one third of primary care patients with persistent symptom diagnoses no longer experienced the symptom(s), according to the survey response. When compared to the level of burden in other primary care patients, persistent symptoms were significantly associated with higher levels of patient burden, including increased severity of symptoms, anxiety and depression, and decreased physical functioning. Based on our findings, we recommend that the burden experienced by patients with persistent symptoms should not be underestimated in clinical consultations and greater awareness of this burden can help improve person-centered care.

4. Clinical prediction models
In Chapter 7, we developed a prediction model for identifying symptom diagnoses that persist for over a year, using data from the Family Medicine Network (FaMe-Net) database. The model was then externally validated with the AHON primary care registry. To facilitate clinical use, we selected predictors that are easily accessible to GPs in daily practice. While the original model showed only marginally acceptable performance, it performed poorly when validated with external data and could not be implemented in clinical practice. However, we identified three robust predictors for developing persistent symptoms namely older age, more previous contacts with the GP and more previous symptom diagnoses. In Chapter 8, we conducted a systematic review to evaluate the quality of AI-based prediction models using electronic health records (EHR) data in primary care, given recent advancements in AI. Out of 15 studies identified, only two were tested in primary care settings. Using the PROBAST guidelines, we found that all studies had an unclear to high risk of bias. As a result, current AI-based prediction models using EHR data are not yet suitable for implementation in everyday primary care practice. Current primary care AI-based prediction models using EHRs available in the literature are not yet suitable for primary care.

Conclusion: Symptom diagnoses are highly prevalent and predominantly managed within primary care settings. The findings of this thesis have several implications for future epidemiological research and clinical practice. For epidemiological research, we demonstrated that utilizing a robust and well-structured primary care EHR database proved to be beneficial for studying the epidemiology of symptom diagnoses. Given the inconsistent definitions of ‘persistent’ symptoms, we recommend a combination of data-driven and experts opinions methods to accurately define the duration of persistent symptoms for future research. From clinical practice, the three robust predictors could be used in daily general practice as an orientation to early identify patients with high risk of developing persistent symptoms. Given the high level of burden experienced by patients having symptom diagnoses that persist for more than a year, we call for greater efforts to a wider implementation of effective interventions for managing patients with symptom diagnoses, particularly those with persistent symptoms. Although promising new interventions are emerging in the literature, further research is needed to validate them before broader implementation. Notably, more education and training for GPs on managing persistent symptoms could enhance care quality.