Introducing Edna: A trainee chatbot designed to support communication about additional (secondary) genomic findings

https://doi.org/10.1016/j.pec.2020.11.007Get rights and content

Highlights

  • Analysis of genetic counseling transcripts provided insights into patient needs.

  • A chatbot can assist patients to make informed decisions for additional findings.

  • Natural language processing algorithms applied to program a digital genetic resource.

  • Chatbot able to collect family history and refer patients to genetic counselors.

  • Conversation agent able to emulate natural flow of genetic counseling sessions.

Abstract

Objective

To support informed decision-making about reanalysis of clinical genomic data for risk of preventable conditions (‘additional findings’) by developing a chatbot (electronic genetic resource, ‘eDNA’).

Methods

Interactions in pre-test genetic counseling sessions (13.5 h) about additional findings were characterized using proponent, thematic and semantic analyses of transcripts. We then wrote interfaces to draw supplementary data from external genetics applications. To create Edna, this content was programmed using a chatbot framework which interacts with patients via speech-to-text.

Results

Conditions, terms, explanations of concepts, and key factors to consider in decision making were all encoded into chatbot conversations emulating counseling session flows. Patient agency can be enhanced by prompted consideration of the personal and familial implications of testing. Similarly, health literacy can be broadened through explanation of genetic conditions and terminology. Novel aspects include sentiment analysis and collection of family history. Medical advice and the impact of existing genetic conditions were deemed inappropriate for inclusion.

Conclusion

Edna’s successful development represents a movement towards accessible, acceptable and well-supported digital health processes for patients to make informed decisions for additional findings.

Practice implications

Edna complements genetic counseling by collecting and providing genomic information before or after pre-test consultations.

Introduction

With the advent of digital health, there is growing interest in the role of electronic tools to support informed decision making. Conversational agents, or ‘chatbots’, that interact with the user via speech or text, are one such tool. Chatbots can be accessed by patients at any time from their own devices and are considered to be an acceptable means of support [1]. People have been shown to interact with chatbots in a similar way as with humans, disclosing both factual and emotional information [2]. Chatbots have been used to deliver text-based information to support consent and participation in a community health research initiative testing for future risk of disease [3].

Chatbots could potentially assist with a highly contentious area of genomic medicine: provision of additional (secondary) findings to patients undergoing clinical genomic sequencing tests.

Additional findings are the result of a deliberate search for disease-causing gene variants unrelated to the medical indication for a genomic sequencing test [4,5]. This is possible because genomic sequencing creates data on thousands of genes, not only those known to cause the patient’s presenting condition. The American College of Genetics and Genomics has recommended that clinical testing laboratories should search for additional findings in 59 genes which cause conditions that are potentially preventable or treatable [4,6]. This caused a storm of debate internationally. Although the availability of additional findings is left to the discretion of the clinical laboratories in some countries [7], currently most national clinical guidelines outside the USA do not encourage making them available [8]. In Australia, laboratories do not generally offer additional findings [7] but action to develop policy on this topic is planned [8]. As national sequencing studies around the world begin to investigate additional findings, policy views may evolve [9,10].

International genetic testing recommendations emphasise that those considering testing must be provided with relevant information to enable them to give informed consent [11,12] and the importance of this has been reiterated in debate regarding additional findings. Where additional findings are offered, a patient makes a decision about additional findings at the same time as their decision to proceed with genomic sequencing test for diagnosis of their condition [13]. There has not yet been a systematic exploration of this process with respect to informed consent, but there are some indications in the literature that people accept, despite giving little consideration to additional findings when having diagnostic testing [14] and hence are not making informed decisions.

To investigate alternative informed decision-making processes, we piloted a different approach [15]. We gave patients the opportunity to decide about additional findings after completing their diagnostic testing. To support informed decision making and adjustment, interested patients received a decision tool and face-to-face genetic counseling. The outcomes of this approach are currently being evaluated; however, this model is clearly labor intensive. As such, it is not feasible for widespread use, as there is a current shortage internationally of genetic counselors [16] and genomic testing has escalated demand for genetic services [17]. Genetic counseling in Australia is predominantly provided through government funded services [18], with funding constraining availability.

Patients are interested in receiving additional findings [19,20], but a delivery model is needed which supports informed decision making by patients, while minimizing the impact on health services. Chatbots have functionality that has the potential to meet this need [3]. As a digital service, chatbots can be accessed at any time to answer general questions. Because they are an artificial intelligence agent, there is no perception of judgement on the nature of questions, so people can ask extremely personal or seemingly ‘naïve’ questions. However, the absence of human contact can be a limiting factor in chatbot uptake, particularly in healthcare, with qualitative evidence suggesting that people are concerned that chatbots lack empathy and are unable to understand emotional issues [21]. This user hesitancy exists even though chatbots are able to identify when interpersonal interaction is required and can be designed to triage people for health professional follow up [22].

A chatbot simulates human conversation through artificial intelligence. In order to recognise the content in the human speech, and provide a meaningful response, it requires a large body of relevant data on which to draw, called the chatbot ‘brain’. Chatbot acceptability is largely dependent on the quality of provided content, with 93 % of participants in one study agreeing that information needed to be accurate and reliable [21]. Programming the chatbot brain with trustworthy, applicable data that can be delivered in a manner that fosters informed, rational decision making is therefore integral to maximising engagement and minimising user hesitancy.

In the first instance, content needs to be gathered from an appropriate source. Many chatbot brains are developed from open source data [23], but such sources are inadequate for the development of chatbots for specific or complex healthcare applications, such as patient decision-making about additional findings. Ideally, the data informing chatbot development needs to be derived from the specific context, in this case, pre-test genetic counseling. Over time, chatbots can facilitate gathering data on frequently asked questions, informing ongoing development of resources and processes which best support informed decision making.

This project describes the development of a genomics chatbot, Edna, using transcript analysis of pre-test genetic counseling sessions for additional findings. We intended that Edna would complement genetic counseling by collecting and providing genomic information in response to patient questions, prior to, following, or in some cases in place of, counseling sessions. Edna was designed to address repetitive or predictable aspects of genetic counseling, allowing genetic counselors to focus on specific issues salient to those clients who choose to have counseling. Edna was not designed to provide information tailored to the specific medical, genetic or social circumstances of the user.

Section snippets

Methods

Ethics approval was obtained under an amendment from the Melbourne Health Human Research Ethics Committee ‘Integration of Genomic sequencing into clinical care: A demonstration evaluation’ project (HM HREC 13/MH/326), with reciprocal approval from the CSIRO Health and Medical Human Research Ethics Committee (CHMHREC 6/2014).

Results

Using a combination of analyses we determined which genetic counseling interactions could be delivered by Edna, including the factors the patient would need to make an informed decision (Section 3.1). We accessed external databases to provide information on conditions and terms (Section 3.2) and explored the decision-making process in finalising the interaction (Section 3.3). We found that family history, while not always directly related to the decision process around additional findings, were

Discussion

Conversations between genetic counselors and patients allowed us to develop a chatbot prototype, Edna, that could potentially complement genetic counseling by collecting and providing accurate, reliable genomic information independent of a face-to-face session. Proponent, thematic and semantic analyses provided insight into the information exchanged, issues raised during counseling for additional findings, and how these were handled. Through these analyses we were able to determine that

Funding

This study was funded by the State Government of Victoria (Department of Health and Human Services) and the 10 member organisations of the Melbourne Genomics Health Alliance.

Work undertaken at the Murdoch Children’s Research Institute was supported by the Victorian Government’s Operational Infrastructure Support Program.

CRediT authorship contribution statement

David Ireland: Methodology, Formal analysis, Software, Writing - review & editing. DanaKai Bradford: Conceptualization, Methodology, Formal analysis, Writing - original draft, Writing - review & editing. Emma Szepe: Software. Ella Lynch: Supervision, Software, Investigation, Writing - review & editing. Melissa Martyn: Investigation, Writing - review & editing. David Hansen: Supervision, Funding acquisition, Writing - review & editing. Clara Gaff: Conceptualization, Funding acquisition,

Acknowledgements

The authors would like to thank the patients and genetic counselors who agreed to share the content of their counseling sessions to assist in creating Edna.

We are grateful to Dr Jill Freyne for insightful comments on the manuscript and Dr Ling Lee for assisting with literature searching and data extraction. Anaita Kanga-Parabia, Callum McEwan and Rigan Tytherleigh reviewed genetic counselor patient notes and/or checked the accuracy of transcripts. Kristina Hood provided invaluable

References (41)

  • Implementation Plan – National Health Genomics Policy Framework 2018-21

    (2020)
  • European Society of Human Genetics

    Opportunistic Genomic Screening. Recommendations of the European Society of Human Genetics (draft Report)

    (2020)
  • M. Morrison et al.

    Implications of secondary findings for clinical contexts

  • Council of Europe

    Additional protocol to the convention on human rights and biomedicine concerning genetic testing for health purposes. Council of Europe treaty series 203

    Eur. J. Health Law

    (2008)
  • Organisation of Economic Co-operation and Development

    OECD Guidelines for Quality Assurance in Molecular Genetic Testing

    (2007)
  • D.F. Vears et al.

    How do consent forms for diagnostic high-throughput sequencing address unsolicited and secondary findings? A content analysis

    Clin. Genet.

    (2018)
  • B.A. Bernhardt et al.

    Experiences with obtaining informed consent for genomic sequencing

    Am. J. Med. Genet. A

    (2015)
  • M. Martyn et al.

    A novel approach to offering additional genomic findings – a protocol to test a two-step approach in the healthcare system

    J. Genet. Couns.

    (2019)
  • J.M. Hoskovec et al.

    Projecting the supply and demand for certified genetic counsellors: a workforce study

    J. Genet. Couns.

    (2018)
  • A. Nisselle et al.

    Australian genomics workforce & education working group

    J. Genet. Couns.

    (2019)
  • Cited by (15)

    • Digital health-enabled genomics: Opportunities and challenges

      2022, American Journal of Human Genetics
      Citation Excerpt :

      Back-end analytics can also be accessed to measure utilization, and chatbots can also be integrated into the electronic health record (EHR) for both automation and documentation. In genetics service delivery, chatbots are being developed and utilized to collect family health history,37 identify individuals at increased risk for hereditary cancer syndromes,38,39 provide education about genetic conditions,40 and support informed decision-making regarding receipt of secondary genomic findings,41 among other use cases. An overarching goal of these chatbots is to free genetic counselors’ and GC assistants’ time for higher-level patient care duties by having the chatbots serve as virtual GC assistants.

    View all citing articles on Scopus
    View full text