All medical decisions start by understanding the patient, their behavior, history, personality and needs, and of course, their symptoms. But compared to the hard science imaging and molecular tests we have. The tools we use for understanding patients are so crude and subjective for our modern data age that they are frankly…a joke.

What am I referring to? Forms and questionnaires.

There is a segment of the medical research community that has poured great effort into making endless different forms and questionnaires. The field calls these “diagnostic tools” and relishes the opportunity to copyright them and prevent others from freely using them. That is all fine until you do one and realize they are ridiculously simplistic.

For example, feel free to diagnose yourself with depression using the following clinically validated diagnostic tool called the PHQ-2 using these diagnostic questions

The PHQ-2

1) "Over the past two weeks, how often have you been bothered by feeling down, depressed, or hopeless?"

a) Not at all (0 points)

b) Several days (1 point)

c) More than half the days (2 points)

d) Nearly every day (3 points)

2) "Over the past two weeks, how often have you been bothered by having little interest or pleasure in doing things?"

a) Not at all (0 points)

b) Several days (1 point)

c) More than half the days (2 points)

d) Nearly every day (3 points)

Alternatively, use the famous and baffling pain scale diagnostic tool:

"On a scale of 0 to 10, where 0 means 'no pain' and 10 means 'the worst pain imaginable,' how would you rate your pain right now?"

Can someone really copyright protect basic questions and sue a researcher/company for infringing on their copyright by asking patients “if they have been bothered by depression in the past 2 weeks?” - No, these are not defensible inventions.

Are answers to these simplistic and broad questions sufficient for precision care? No way!!

There are many different subtypes and patterns of mental illnesses that we don’t understand because we use these crude approaches. Pain patterns are rich with information that is not captured or understood. There are different patterns and types of fatigue. Flu fatigue has a different pattern from chemotherapy-induced fatigue or exercise fatigue. Nonetheless, there are literally thousands of these types of simplistic questionnaires, and they are key in medicine and in research into patient symptoms, behavior, history, mental health, quality of life, outcomes, drug responses, and much more.

We all know this is too simplistic and crude for precision medicine. Patients dread filling out all these questionnaires and speed through them as quickly as possible. But, what is the alternative?

Well, in the modern world of artificial intelligence, questionnaires are dead because we can use technology to objectively understand and measure behavior and symptoms….and that is important. I will post more about these solutions in the future, but let’s focus on the need here.

Questionnaires revealed the great potential for precision medicine and AI technologies in scalable, deep patient phenotyping by analyzing numerous objective measurements of patient behavior and symptoms using a smartphone.

Patient Reported Outcome (PRO) Questionnaires show promise for precision medicine that could be vastly improved with new AI-driven technologies. I believe that the only viable platform for scalable precision medicine is the smartphone – a topic for another post. Let’s discuss the need and applications in oncology.

A meta-analysis of studies of various solid tumor types showed that subjective PROs are significantly associated with objective radiographic measures of tumor responses to treatment 1. Nearly all patients showing a partial response (PR) or stable disease (SD) status had improved PRO scores, whereas those with progressive disease (PD) reported a decline. PRO response duration was shorter than radiographically measured tumor response durations, suggesting that the patient might detect worsening condition before oncologists can detect tumor growth. Indeed, cancer symptom responses have been shown to be meaningful assessments of early cancer treatment effects and early signals of progression 2,3. Thus, PROs are informative endpoints that could inform earlier treatment changes.

PROs also predict clinical outcomes, including hospitalization and survival 4–7. Additionally, symptom clusters defined from PROs and the interconnections between diverse symptoms (eg. fatigue, pain, nausea, skin rash, etc.) in a cluster have been shown to depend upon the cancer type, revealing specificity 8. Different symptom clusters link to different biological mechanisms 9. Symptom cluster network structures derived from 38 measures were also shown to evolve overtime, illuminating “sentinel” symptoms for monitoring disease activity, and predicting outcomes in breast cancer 8. Thus, PRO studies point to the potential for symptom phenotyping to guide adaptive treatment decisions. Symptom phenotypes differ between patients and can delineate important subgroups 10.

The above studies show that even crude and subjective PRO questionnaires yield valuable data for monitoring treatment responses. We therefore expect that vastly improved technology solutions for deep AI-supported symptom and behavior phenotyping will substantially advance capabilities for adaptive care models, as described below.

Problematically, subjective questionnaires are a barrier to precision medicine and drug development in oncology.

While changes to cancer symptoms are linked to important outcomes, existing data capture solutions are simplistic PRO questionnaires that are not ideal for precision medicine applications that require accurate and objective data. PROs in open-label trials, where patients are aware of their treatment, suffer from problematic biases because the patient’s perception of their symptoms or function is influenced by knowledge of their assigned treatment 13. For example, a large meta-analysis of 1346 trials across different therapeutic areas found that treatment effects for “subjective” outcomes (ie. clinician-reported outcomes and PROs) were larger in open-label trials than in blinded trials, suggesting bias 14. Study designs incorporating more objective measures showed little evidence of bias compared to those with subjective measures 15. Thus, incorporating objective measures is important.

Concerns of bias with subjective PROs in open-label trials have also been highlighted by the FDA and have been problematic for FDA interpretations of some endpoints in oncology trials 16. Consequently, although it is a new area for development, the field is pushing for more objective measures 17–19. Important lessons are also available from the mental health field, where subjective and overly broad DSM-V diagnostic tools were identified as a major barrier to precision medicine and the discovery of underlying mechanisms of mental illness, which resulted in fundamental changes to research methods enforced at the NIH level 20–22.

Overall, patient symptom analysis is critical to care decisions, but a shift to more objective, data-driven, artificial intelligence (AI)-based approaches for measuring patient symptoms and behavior would better serve precision medicine applications and improved cancer care models 6-8. 9.

There is a very important and exciting new application for AI that, unlike questionnaires, is defensible and can therefore create a new generation of precision medicine technology solutions.

 

References

1. Victorson, D., Soni, M. & Cella, D. Metaanalysis of the correlation between radiographic tumor response and patient‐reported outcomes. Cancer 106, 494–504 (2006).

2. Secord, A. A. et al. Patient-reported outcomes as end points and outcome indicators in solid tumours. Nat. Rev. Clin. Oncol. 12, 358–370 (2015).

3. Bouchard, L. C., Aaronson, N., Gondek, K. & Cella, D. Cancer symptom response as an oncology clinical trial end point. Expert Rev. Qual. Life Cancer Care 3, 1–12 (2018).

4. Papachristou, N. et al. Network Analysis of the Multidimensional Symptom Experience of Oncology. Sci. Rep. 9, 2258 (2019).

5. Miaskowski, C. et al. Advancing Symptom Science Through Symptom Cluster Research: Expert Panel Proceedings and Recommendations. J. Natl. Cancer Inst. 109, djw253 (2017).

6. Sullivan, C. W., Leutwyler, H., Dunn, L. B. & Miaskowski, C. A review of the literature on symptom clusters in studies that included oncology patients receiving primary or adjuvant chemotherapy. J. Clin. Nurs. 27, 516–545 (2018).

7. Kerrigan, K. et al. Prognostic Significance of Patient-Reported Outcomes in Cancer. JCO Oncol. Pr. 16, e313–e323 (2020).

8. Kalantari, E., Kouchaki, S., Miaskowski, C., Kober, K. & Barnaghi, P. Network analysis to identify symptoms clusters and temporal interconnections in oncology patients. Sci. Rep. 12, 17052 (2022).

9. Lu, K. et al. Integrated network analysis of symptom clusters across disease conditions. J. Biomed. Inform. 107, 103482 (2020).

10. Miaskowski, C. et al. Latent Class Analysis Reveals Distinct Subgroups of Patients Based on Symptom Occurrence and Demographic and Clinical Characteristics. J. Pain Symptom Manag. 50, 28–37 (2015).

11. Wilson, M. K., Karakasis, K. & Oza, A. M. Outcomes and endpoints in trials of cancer treatment: the past, present, and future. Lancet Oncol. 16, e32–e42 (2015).

12. Burris, H. A. et al. Improvements in Survival and Clinical Benefit With Gemcitabine as First-Line Therapy for Patients With Advanced Pancreas Cancer: A Randomized Trial. J. Clin. Oncol. 41, 5482–5492 (2023).

13. Roydhouse, J. K., Fiero, M. H. & Kluetz, P. G. Investigating Potential Bias in Patient-Reported Outcomes in Open-label Cancer Trials. JAMA Oncol. 5, 457–458 (2019).

14. Wood, L. et al. Empirical evidence of bias in treatment effect estimates in controlled trials with different interventions and outcomes: meta-epidemiological study. BMJ 336, 601 (2008).

15. Savović, J. et al. Influence of Reported Study Design Characteristics on Intervention Effect Estimates From Randomized, Controlled Trials. Ann. Intern. Med. 157, 429 (2012).

16. Gnanasakthy, A., Barrett, A., Evans, E., D’Alessio, D. & Romano, C. (DeMuro). A Review of Patient-Reported Outcomes Labeling for Oncology Drugs Approved by the FDA and the EMA (2012-2016). Value Heal. 22, 203–209 (2019).

17. Jahedi, S. & Méndez, F. On the advantages and disadvantages of subjective measures. J. Econ. Behav. Organ. 98, 97–114 (2014).

18. Moustgaard, H., Bello, S., Miller, F. G. & Hróbjartsson, A. Subjective and objective outcomes in randomized clinical trials: definitions differed in methods publications and were often absent from trial reports. J. Clin. Epidemiology 67, 1327–1334 (2014).

19. Roach, K. E. Measurement of Health Outcomes: Reliability, Validity and Responsiveness. JPO J. Prosthet. Orthot. 18, P8–P12 (2006).

20. Insel, T. R. & Cuthbert, B. N. Brain disorders? Precisely. Science 348, 499–500 (2015).

21. Cuthbert, B. N. & Insel, T. R. Toward the future of psychiatric diagnosis: the seven pillars of RDoC. BMC Medicine (2013).

22. Sanislow, C. A., Ferrante, M., Pacheco, J., Rudorfer, M. V. & Morris, S. E. Advancing Translational Research Using NIMH Research Domain Criteria and Computational Methods. Neuron 101, 779–782 (2019).