Analysing Free-Text Patient Reports to Identify Key Symptoms in Parkinson’s Disease

Listening Beyond the Questionnaire

The experiences of people living with Parkinson’s disease are often captured using check-box style surveys, which, while practical, cannot fully reflect the complexity of their daily challenges. The PD-PROP addresses this issue by allowing patients to describe, in their own words, the problems that affect them most. This approach provides an authentic view of life with Parkinson’s disease, capturing common symptoms such as tremor as well as under-recognised problems such as fatigue, anxiety, or fear of falling.

Marras et al., aimed to analyse these detailed, free-text responses on a scale that would not be possible through human review alone. They developed a “human-in-the-loop” process that combined expert interpretation with natural language processing and machine learning. The method was applied to 168,260 responses from approximately 25,000 patients with PD. Nine curators, including clinicians, people with PD, and a non-clinician PD expert, worked together to define and classify symptoms into clinically relevant categories.

The method achieved an accuracy of 95% in validation tests and generated data that preserved the patient’s voice while providing clear, clinically relevant insights for treatment and research planning. The most frequently reported initial problems were tremor, gait difficulties, and pain or discomfort, providing a clear picture of the challenges most often prioritised by patients.

Mapping the Full Spectrum of Patient Experiences

Through this combined human and machine approach, 65 distinct symptoms were identified and grouped into 14 domains. These domains included motor problems such as tremor, rigidity, bradykinesia, postural instability, gait disturbances, and other motor symptoms. Also included were non-motor categories such as sleep, fatigue, cognition, psychological issues, pain, autonomic dysfunction and dyskinesias.

Initially, 46% of patients identified tremor as a major problem, 39% reported gait issues not otherwise specified, and 33% described experiencing pain or discomfort. Other common problems included impaired dexterity or micrographia (31%), balance issues (29%), physical fatigue (29%), and speech difficulties (23%). Please refer to Table 4 in the manuscript for the complete breakdown of symptom frequencies.

Several patterns emerged across demographic groups. Women reported a higher frequency of symptoms in nearly all domains, with the greatest differences seen in psychological problems, pain, fluctuations, and dyskinesias. Older adults, particularly those over 60 years of age, were more likely to report issues with gait, balance, and autonomic function. Men over 60 years old reported cognitive difficulties more often, while women in the same age group reported fatigue more frequently. The curated dataset captured both the diversity and prevalence of symptoms across the PD population. It also provided a foundation for tracking changes over time and exploring associations with demographic and clinical characteristics.

Not so fast

Although the PD-PROP approach, strengthened by human expert curation, successfully captured a wide range of patient-reported problems, the study had some limitations. Responses could only be entered via keyboard, which may have limited participation for those with dexterity challenges, potentially reducing representation of certain groups. The classification of responses relied on curator interpretation, which could not be personally verified with the patients and carried a risk of misinterpretation. The curated symptom list was influenced by the prior knowledge and experience of the curation team, which could have led to underrepresentation of less-recognised symptoms. In some cases, the limited detail in the responses made it difficult to distinguish between related symptoms, requiring the merging of specific categories. Finally, despite reviewing many verbatim reports, it is possible that some problems expressed by patients were not captured.

Why does it matter?

Parkinson’s disease affects millions of people worldwide, yet traditional symptom measurement tools often limit responses to predefined checklists. This research presents a scalable method for understanding the symptoms that patients consider most important, as expressed in their own words. In this large PD-PROP dataset, tremor was reported by 46%, gait issues by 39%, and pain by 33% of respondents at their first report. Such insights reveal patterns that vary by age and sex, providing a nuanced understanding into the lived experience of PD. These findings can inform clinical priorities, shape research agendas, and aid in selecting meaningful outcome measures for future trials, ensuring that patient perspectives are central to decision-making.

Take home messages

1. Combining expert curation with natural language processing and machine learning can accurately and efficiently facilitate analyses of large-scale free-text patient reports.
2. The most frequently reported initial symptoms in PD were tremor, gait difficulties, and pain, with patterns varying across age and sex.
3. This approach offers a powerful tool for aligning research and clinical care with the symptoms that matter most to people living with PD.

Guest author: Hassan Khan, MSc.

Reviewer: Barbara Fahmy, MS OTR, MPA

This article was written as part of a series of ‘journal club’ summaries for Scientific Writers Ltd and is based on the following publication.

Title: What Patients Say: Large-Scale Analyses of Replies to the Parkinson’s Disease Patient Report of Problems (PD-PROP)

First Author: Marras C, et al.

Journal: Journal of Parkinson’s Disease

Date online: 09 June 2023