ChMP: A Collaborative Medical History Portal

Abstract

Family medical histories play an invaluable role in disease prevention, diagnosis and treatment. Self reported medical histories frequently contain incorrect or incomplete information, severely diminishing the quality of care and clinical outcome of the patient. While tools for obtaining and analyzing medical histories are available to medical professionals, no system exists to allow families to actively participate in the collection and utilization of medical history data. We have developed a free web-based service (http://www.inherithealth.com) that allows a family to collaboratively capture and store medical history information relevant to breast cancer. The service is built on a custom framework that enables the integration of existing breast cancer risk assessment models with web-based software to communicate evidence-based risk assessment to consumers. Preliminary user evaluations indicate that consumers find the tool usable, and are interested in learning about their breast cancer risk.

Introduction

Family medical history is an important tool used by clinicians to assess an individual’s risk of disease. It has been shown that family medical histories can be valuable in prevention, screening and diagnosis of disorders such as heart disease, colorectal cancer, diabetes, and breast cancer.

After age, family medical history is the second most important risk factor for predicting breast cancer in women¹. Risk is dependent on a variety of features including the number of relatives affected with related cancers and the age at which the relatives developed cancer².

The importance of related cancers in family members for predicting a patient’s disease risk emphasizes the need for a complete and accurate family medical history in medical records. The primary issues preventing the consistent use of family medical history by primary care physicians is underestimation of the utility of family medical histories for diagnosis and treatment, and the amount of time and information required to collect and organize histories³. The present study aims to demonstrate that complex breast cancer risk models based on family medical history can be made accessible to healthcare consumers. We propose a tool that enables consumers to collect and store their family medical history, shifting the task out of the clinical setting. In addition to saving physician time and encouraging collaboration, we show that the collection of family medical history in a structured format can be used to provide high-level risk assessment to health care consumers.

Background

Guidelines are available that describe best practices for constructing a family medical history^4–6 but there are very few tools designed for non-medical professionals to build and populate a family medical history. Internet tools such as the American Medical Association’s Adult family history form⁷ and the Health and Human Services family history website⁸ enable users to collect, organize and store their family medical history, but do not support collaboration by multiple family members, nor do they use the information to provide feedback to the individual. Disease specific questionnaires such as the Myriad hereditary cancer quiz⁹ and the National Cancer Institute’s Risk Assessment Tool¹⁰ provide a risk assessment for breast cancer, but do not take a full family medical history into account.

Most tools that are available for the acquisition of family medical records and risk assessment for breast cancer are geared towards medical professionals, including genetic counselors and medical geneticists. CancerGene¹¹, Progeny¹², and Cyrillic¹³ are software packages commonly used by genetic counselors. Each requires domain knowledge about pedigree drawing, familiarity with medical terminology and detailed knowledge about breast cancer risk.

Cyrillic and CancerGene are widely used by genetic counselors to compute a patient’s risk of breast cancer using peer-accepted and validated statistical models. These models, such as Gail¹⁴, Claus¹⁵, BRCAPRO¹⁶, and Tyrer-Cuzick¹⁷, are derived from longitudinal studies of large populations in which disease outcomes are statistically associated with a set of patient attributes.

Methods

We have developed a free web application that integrates family medical histories and breast cancer risk assessment models to provide quantitative and qualitative risk assessment to individuals in a family. The application is designed to support collaboration by multiple family members. The framework supports the integration of newly developed models to provide the most current risk assessment tools that become available in the public domain.

The web-based interface enables multiple users to generate and modify a family tree using an original pedigree drawing algorithm. The tree drawing algorithm ensures node spacing that is consistent with a hand-drawn pedigree, including heuristics to ensure maternal/paternal tree balancing. When a new node is added, the algorithm attempts to minimize the effect of the addition on the overall tree structure.

As the number of relationships in a pedigree increases, filtering the relevant relationships for a particular individual becomes necessary to emphasize the relations of importance. The user can change the proband of interest, causing the tree to be redrawn and centered with the new proband as the focus. A zooming user interface is used to support visualization at multiple resolutions. As the tree becomes more complex, zooming provides an intuitive means for navigation by maintaining a consistent context¹⁸.

The database schema for the application is based on the Genealogical Data Communication (GEDCOM) format. Entries within the database are consistent with the concept of the nuclear family. The degree of relationship to the proband can be inferred from this structure. Database functionality is implemented with MySQL using Ruby on Rails. Users are required to register for the service by providing a verifiable e-mail address. After a secure login, the user can build the structure of the family by adding parental, sibling, spousal, and child relationships. Modifications can be made to the individual nodes in the tree, but not to the structure of an existing tree.

Collaborative Tree Building

The creator of the tree can grant permission to other users to view and modify the tree. To share an instance of a family tree, a user enters the email address of the individual with whom they wish to share. The share action is handled by the controller and produces two e-mails, one to the initiator of the share and the second to the new user. The e-mail to the new user notifies them that the sender has shared their family health history, and contains a server generated temporary password. A second e-mail is sent to the initiator of the share, instructing them to send an e-mail to the new user containing another system generated password. In order to authenticate, the new user must supply both the system generated passwords. The dual passwords provide redundancy to maintain the privacy of sensitive medical information.

Node editing concurrency is handled via locking on a per node basis. Multiple users can work on the family tree simultaneously, but cannot edit information for the same node.

Medical Information Acquisition

In the current version of the software, only medical information required by the breast cancer risk assessment models is collected. ChMP uses a controlled terminology based on ICD-10 that collapses common terms in a hierarchy into a single generic diagnosis to capture medical information. The terminology for ChMP only contains diagnoses relevant to the breast cancer risk assessment models integrated with the system. The software uses the terminology in conjunction with common user interface idioms (i.e. auto-complete, drop-down) to encourage the user to supply information that will help inform the specific models, and limit users from entering other medical conditions.

Once the tree is populated with family medical history data, the user has the option to use this information to calculate relative risk of acquiring breast cancer.

Risk models

Two models, BRCAPRO¹⁹ and Gail²⁰, were adopted from the public domain to compute risk assessment for breast cancer. The Claus model was implemented as described in [15] using Java.

Each of the models requires different attributes of the individual to calculate risk. Figure 1 shows the specific input requirements for the three models currently used in ChMP, and a fourth commonly used model (Tyrer-Cuzick).

Figure 1.

Breast cancer risk model parameters

Risk assessment framework

All risk assessment models that utilize family history data require traversal over the family tree to aggregate information. The risk assessment framework allows for the abstraction of generic tree traversal methods while enabling specific data retrieval relevant to each model.

The risk assessment model integration framework consists of three components: the query interface, model specific wrappers, and the risk assessment results. The framework and model wrappers are implemented in Java, and communicate via XML-RPC with a ruby client.

The query interface allows specific models to query the details of a family medical history in an abstract manner. It provides a generic set of methods that the model wrapper can call to access the information required by that model. The interface implementation translates the query into model specific formats. This approach provides a layer of abstraction over the details of tree search to support rapid integration of new models, all of which require similar methods.

The interface specifies the syntax and semantics of the input and output of a query. It is neutral with respect to the particulars of the back-end. As such, a query could be processed by accessing the results from an XML file in the same way it would access a database backend or a web service. The query interface is used by the individual models to access the underlying family medical information.

Each model is encapsulated by a wrapper class responsible for translating generic results of the query into the specific format required by the underlying model. These classes make requests to the family tree using methods provided by the query interface. Query results are parsed and the parameters are passed to the appropriate methods of the model. Results of the risk assessment model are output in a common results format for processing by the client application.

The generic risk assessment results interface provides a method to access the quantitative results of the model query, and any details or exceptions resulting from the calculations. All model wrappers implement a common results interface, which can be extended to provide additional model-specific results.

Usability evaluation

Six users were asked to complete 5 tasks using ChMP. The users (2 female, 4 male) ranged in age between 23 and 59. The tasks cover many of the features available in ChMP including construction of a family tree, entry of personal and medical information, and generation of a breast cancer risk assessment. Users were provided with two pieces of documentation to complete the tasks: a short (< 3min) introductory video²¹ that demonstrates how to perform basic functions in the ChMP user interface and written instructions provided with the tasks that explain how to use the required functions that are not covered in the video. The developers did not assist or interact with the testers while they completed the evaluation.

After completing the tasks, users were asked to provide quantitative and qualitative feedback about their experience using the system.

Results

The model implementations were validated by comparing the risk assessment results with CancerGene¹¹ results for equivalent pedigrees. We used an abbreviated pedigree from the literature¹⁶ and a pedigree supplied by a counselor at the Stanford Cancer Genome center as input to both systems to validate the models. We selected two probands from each of the pedigrees and computed the lifetime (to age 79) risk of breast cancer.

ChMP utilizes the same low-level model code as BRCAPRO; as such, we did not observe any discrepancies in results from the two applications. The results of the Gail model are slightly different than those generated by CancerGene because CancerGene uses a different data set to compute prior probabilities. We expected little or no difference between the results of the Claus model.

Discrepancies between the results were minor [Table 1]. There were no differences between the BRCAPRO and Claus models under all tests. We account for the discrepancy in the Gail models due to the different model generating dataset for each. The ChMP implementation of Gail produced identical results to the NCI’s online tool¹⁰.

Table 1.

Model Validation Test Results: lifetime risk assessment outputs for each pedigree, system, and model. “Jane” is a 48-year-old female with 3 first-degree relatives with early onset of breast cancer. “Betty” is a 68-year-old female with 2 first-degree relatives with late onset of breast cancer. “Lucy” is a 65-year-old female with a first- and second-degree relative with breast cancer. “Cindy” is a 58-year-old female also with a first- and second-degree relative with breast cancer and a first degree relative with ovarian cancer.

	Claus		Gail		BrcaPro
	CG	ChMP	CG	ChMP	CG	ChMP
Pedigree 1
Jane	0.35	0.35	0.29	0.26	0.21	0.21
Betty	0.03	0.03	0.13	0.15	0.07	0.07*
Pedigree 2
Lucy	0.05	0.05	0.07	0.08	0.08	0.08
Cindy	0.08	0.08	0.12	0.1	0.09	0.09*

^*Estimated at age 78.

Usability evaluation

Results from the usability evaluation are summarized in Table 2. All of the testers completed the tasks correctly as indicated in the evaluation instructions. The five tasks took an average of nine and a half minutes to complete. Overall, the testers found the navigation and data entry to be intuitive and easy to use. One user reported that “you don’t need to be computer savvy to figure out the gist of it fairly quickly.” The chief complaint regarding system usability was that the system felt “brittle and clunky.” When asked about the risk assessment feature, one user reported that the text was unclear and that the graphs were difficult to interpret. The physician evaluator suggested that a biostatistics degree may be required to interpret the risk assessment results. However the physician also reported that it would be useful with training and expressed interest in using it with his patients.

Table 2.

Mean score (SD) for survey questions. 1-strongly disliked, 5-strongly liked.

navigation	data entry	overall experience	value of risk as presented is useful
4.5 (0.5)	4.7 (0.4)	4.3 (0.4)	3.8 (1.3)

Coordinating collaborative testing proved difficult and outside the scope of the current study. In lieu of user testing for these features, the authors validated the collaborative functionality: sharing a family tree and concurrency locking on individual nodes. Both features operated as expected. The authors do note that the redundant authentication for sharing via multiple passwords posed a significant tradeoff between ease of use and perceived security.

Conclusion

We have developed a web-based application in which users can create and modify their family medical histories. These structured medical histories can be used to automatically generate risk assessments for breast cancer based on peer-reviewed models. Preliminary usability evaluations through surveys indicate that users in general find our system to be intuitive and worthwhile. The risk assessments provided are equivalent to those provided by clinically relevant applications such as CancerGene. While the system provides valuable information, there are areas in which improvements are required.

The widespread adoption of any service in which medical information is provided to a third party requires tremendous attention to privacy and security. Our system makes a first pass at some of these concerns by not requiring identifiable information such as name or birth date, and using dual passwords for new user authentication. More extensive surveys are required to assess users’ specific security concerns.

In addition to communicating directly with the user, the system should also communicate relevant information to a clinician. The most natural solution is a direct interaction with the patient’s electronic health record (EHR). However many providers have not adopted EHRs and for those that have, security and HIPAA regulations are prohibitive in the short term. In future work, we will create a detailed “clinician view” of the data containing all of the relevant information that contributed to the risk assessment. The user will be able to print out this report and discuss the results directly with their physician.

ChMP can be beneficial to breast cancer risk assessment in providing patients information that can be used to enhance screenings. ChMP has been show to be useful for breast cancer but there are many other diseases that exist in which risk assessment models are available. These models can be easily incorporated into our framework so that users have a central portal to investigate their risk of multiple diseases for which risk models are publicly available.

Figure 2.

Data flow through the framework. A risk assessment request is initiated by a client application, handled by the query interface, processed by the risk assessment serve, returning multiple risk assessment’s in a common format.