Building a Pan-Canadian Primary Care Sentinel Surveillance Network: Initial Development and Moving Forward

The Need for Primary Care Data Sources in Public Health Surveillance

Although historically used for detecting infectious disease outbreaks¹⁰ and occupational health problems,¹¹ the concept of sentinel surveillance is increasingly applied to the field of chronic disease surveillance.¹² It is based on the concept that one or more sites are chosen to collect clinically verified information (eg, risk factors, diagnosis) about relatively few individuals, representative of a larger population, to identify public health events of interest.

The medical care of people with chronic diseases is, by and large, managed by primary care physicians. Information about these people and their care is mostly locked away in medical records and not a readily available data source for research or surveillance. In addition, patients in primary care often have multiple chronic health conditions and family physicians are their major source of medical care. For example, in the 2007 survey of Canadians’ Experiences with Chronic Illness Care, 63% of patients with 1 or more chronic condition indicated that their primary health provider explained examination results to them and 68% with 1 or more chronic conditions indicated that their primary health provider explained specific test results to them.¹³ Currently, information about the health of Canadians is derived from national databases that provide details about, for example, mortality statistics, hospital discharge data, disease specific registries, or national population health surveys. However, some health conditions do not always lead to death or generate hospital admissions. Although self-reported survey data can provide useful information about chronic diseases and their risk factors, they are limited because people may not know that they have a specific condition or may not be willing to report it. Further, in Canada there is a single-payer system, and billing data from the provincial health insurance plans are unreliable sources of data. These data sources often do not tell the story of the care of people with chronic disease for many reasons, including the inability to list multiple reasons for the clinical encounter/visit. To produce a more complete picture of the health of the population, clinically verified information collected directly by health care providers is necessary. The data needed, including practice-level data about disease management, both pharmaceutical and nonpharmaceutical, are most readily available where the population receives ongoing health care for chronic diseases. It is clear that primary care practices may provide a rich source of data for chronic disease surveillance if only we can get at it.¹⁴

Chronic Disease Surveillance Objectives

A 2006 study of 980 patients in a family medicine practice in Sherbrooke, Quebec, determined that 89% to 100% had multiple chronic conditions. The mean number of chronic conditions of patients eligible for each applicable condition ranged from 5.5 ± 3.3 to 11.7 ± 5.3.¹⁵ Issues around the morbidity of multiple chronic diseases underscore the need for surveillance of chronic disease based on the following objectives for the CPCSSN feasibility project:

1. Develop an infrastructure for CPCSSN that will underpin the operations of a robust, longitudinal data collection and maintenance of a primary care data repository on chronic disease.

2. Demonstrate the ability to extract relevant data from multiple EMRs in multiple primary care practice sites.

3. Create a usable CPCSSN database that will be a searchable data repository for primary care researchers and will be the basis for reports for government and others about chronic disease in Canada.

Initial Development

The founding networks for CPCSSN were the Atlantic Practice Based Research Network (Memorial University of Newfoundland); Q-NET (Centre santé et des services sociaus de Laval); the Centre for Studies in Primary Care network (Queen's University); the North Toronto Research Network (University of Toronto); the DELPHI network (Thames Valley Research Unit, University of Western Ontario); the Southern Alberta Primary Care Research network (University of Calgary); and the Alberta Family Practice Research Network (University of Alberta). All of these academic primary care research networks had associated family medicine practices that used EMRs. A recent example of an exploration of the process of change to implement clinical guidelines for primary and secondary prevention of cardiovascular disease in primary care practices used a common EMR.¹⁶

To limit the amount of data we would collect, the group decided to focus on 5 chronic diseases: hypertension, diabetes, chronic obstructive lung disease, depression, and osteoarthritis. We also hired consultants to support business and strategic planning and the development of the information technology and data management infrastructure. The initial deliverables for each regional network included the development of a memorandum of agreement (MOA) between the CFPC and the local university or hospital and an ethics application for research ethics board (REB) approval for each site. Once the MOA was signed by all parties, funds flowed to the regional networks as restricted/conditional sub-grants from the CFPC. Funds continue to flow based on deliverables met, in accordance with the main contribution agreement conditions required by PHAC. With the MOAs and REB approvals in place, we then recruited practices that used EMRs. Because of the work involved with assessing data extraction capabilities from individual EMRs, each network was restricted to recruiting up to 10 practices using the same EMR.

Business Planning

Although the initial feasibility project only lasted 7 months, the CPCSSN board of directors decided it was necessary to take a long-term view of the development of the network; therefore a 5-year business plan was developed. As part of this plan, a governance structure, terms of reference for the board and each of the subcommittees, as well as a mission and vision were established. Early in the project we tried to determine who the stakeholders of this network would be and who would be potential consumers of the data repository being developed by CPCSSN. The network is organized as a sub-entity of the CFPC with its own board and a number of sub-committees, as outlined in Figure 1. The board consists of directors of each participating network and representative members of the CFPC. The CPCSSN board worked diligently in phase 1, meeting bimonthly via teleconference and having 2 face to face meetings; the data managers group met every week for 2-hour sessions via teleconference using GoTo Meeting visual technologies (Citrix Online, LLC, Goleta, CA). Funding for CPCSSN is initially through PHAC but future financial sustainability may require funding over and above that expected to come from PHAC. We continue to develop additional funding sources to maintain the infrastructure of the network.

• Download figure
• Open in new tab

Figure 1.

Consensus governance structure for the Canadian Primary Care Sentinel Surveillance Network (CPCSSN).

Human Resources

To develop and maintain a robust longitudinal database, careful consideration was given to the necessary human resources. We had network directors at each site but also decided that a key position was a full-time data manager at each network site. The EMR data are all sent to a regional repository where they are cleaned and stored in preparation for transfer and merging at the central repository. Because each network is working with different EMRs, having someone like a regional data manager, who could become very familiar with that system and who would be able to develop a personal relationship with the sentinel practices, is crucial. Further, having a regional data manager would help to build capacity for additional primary care research at each network. We have also employed a part-time research assistant at each site for regional research development and data reporting. In phase 2 we have established a central repository and hired a senior data manager who will look after the repository and provide assistance to regional data managers. We have also established a central office at the CFPC for the project manager/director and support staff. The chair of the board of CPCSSN is currently at Queen's University, where the central repository is housed. In the future, however, the chair may be at another network so the funding for the office of the chair will float.

Privacy and Security

In the first phase of the project, 6 out of 7 networks obtained REB approval for the project. The ethical issues focused on by the REBs were the security of patient health data, the need for patient consent, and a de-identified database. All of the approvals were based on not requiring individual patient consent for extraction of records. Signs were posted in the waiting rooms of each practice notifying patients that their personal health information is de-identified; if they do not want their health information used, they can request that they be excluded. Although this has not been a problem to date, the process for removing them from the database would happen during the data extraction and cleaning process at the regional network. These patients will have an assigned CPCSSN number that would be flagged for exclusion on data pulls. Only the practice will know the identity of these patients.

In Quebec, REB approval was not granted during the first phase of the project. The committee's major concern was related to the lack of explicit patient consent and, to obtain approval, individual patient consent would be required. This has delayed practice recruitment in Quebec but will allow for measurement of nonparticipation by patients and comparisons of participation rates as well as other differences between Quebec and other provinces.

All identifying patient information will be stripped from the database before it leaves the practice. A CPCSSN number, which consists of a network and site identifier and the unique EMR number, is assigned to each patient. Before data are transferred to the central data repository, each regional network data manager will apply risk identification software to the database to ensure that it is adequately de-identified.

EMRs

During phase 1, 6 proprietary EMRs were used: DaVinc (Montreal, Quebec); Healthscreen (Toronto, Ontario); MedAccess (Kelowna, British Columbia); Nightingale (Nightingale VantageMed Corp, Rancho Cordova, CA); P & P (P & P Data Systems, Inc, Ontario); and Wolf (Wolf Medical Systems, Surrey, British Columbia). All EMRs have different coding structures and each network decided on the best way to access the data for extraction (direct versus frontend). Initial test draws of the patient databases identified what information can be extracted. We are learning how to extract data from these EMR programs but it is clear that, as EMRs increasingly becoming a necessary tool in health care, careful selection of a vendor must also consider needs for data extraction for research.¹⁷

Disease Definitions

Disease definitions have been developed for each chronic disease based on International Statistical Classification of Diseases and Related Health Problems (version 9) codes for diagnosis in combination with certain drugs (from preconstructed drop down lists) and/or positive test results related to the disease. Although these codes lack the specificity preferred for primary care, this is the current coding standard for many EMRs in Canada and will be a potential limitation for the initial data extraction process. The development of metadata and data processes is explained more fully by van Vlyman and de Lusignan¹⁸ in their discussion about a defined number of named elements that convey meaning, given that medical data are complex to process. Access to diagnostic procedures (eg, spirometry) or referrals (eg, referral to psychiatrist) will not be available in all EMRs and will be detailed, as applicable, for each practice site. The “gold standard ” of disease definition is the clinical diagnosis of the primary physician. We will provide each physician with a list of their patients who have been identified from EMR data as an index case with chronic disease(s) and ask them to verify this finding. This will require some initial work by the physician but, with subsequent data draws, the number of patient verifications will be much fewer as will the consistency in EMR data input. CPCSSN disease definitions are included as Appendix 1.

CPCSSN Database Development

An entity relationship diagram has been developed that creates a structure for the types of data that is being collected for the database. Each patient has a unique CPCSSN identifier. General categories of data extraction are network and provider identifiers, patient demographics (de-identified), encounter date and encounter type, health condition, physical examination, risk factors, referrals, laboratory investigations, procedures, and medications.

Data Flow

The CPCSSN data repository is a de-identified database of patients with any of the 5 chronic diseases of interest. All patient identifiers are removed from the database before data leaves the practice and only the de-identified data will be kept on a secure regional network server. Once the data has been cleaned, it will be transferred to the central repository server (Figure 2). Data will be extracted from the EMRs every 3 months during phase 2 operations, for a total of 3 extractions. The first iteration was a test extraction of all available data in the EMRs. The second extraction will be full extraction of all available data in preparation for the third and final extraction, which will be designed to extract new information (different from that extracted in the previous extraction) to form the final data warehouse.

• Download figure
• Open in new tab

Figure 2.

Data repository for regional and central sites at the Queen's University High Performance Virtual Computing Laboratory. System 1, 2, etc represents each regional network site. Connection to the site server will be through a virtual private network.

Recruitment and Retention of Practices

Fundamental to the success of this project is the ability to attract and maintain primary care practices in the networks. Anecdotally, many family physicians with EMRs see—once they get over start-up frustration—the benefits from contributing patient health data for understanding chronic disease in a wider context. All physicians, however, are extremely sensitive to the privacy of their patients and want strong assurances that any health data taken from their electronic record is secure and de-identified. Some remain unconvinced despite these assurances. Other issues are well known.²² We have dealt with the barriers in a variety of ways. We have tried to limit the requirement for physician involvement so that participation will be a minimal burden on the physician or the employees of the practice. Although we do have some practice compensation, this is not a driver for practices to participate. We are developing a process for participating physicians to get CFPC continuing professional development credits and we plan to develop regular reports to provide feedback to sentinel physicians about their practice as well as provincial and national comparators. We want participating practices to be sentinels for many years but at the moment we do not have any estimate about “sentinel fatigue.”

Stakeholder Development

The current major stakeholders working with the network are PHAC, CFPC, and the Canadian Institute for Health Information (CIHI). CIHI has been involved since the early days of the project and, because of their expertise with handling health data, they have been instrumental in the development of data element definitions and data capture processes.

Our relationship with CIHI will eventually allow for linkage studies with other health administrative databases. Developing stakeholder relationships with provincial and national health organizations and health professional data holders is also pivotal to CPCSSN′s success. There may be other groups for whom the network could provide valuable health data, such as for adverse event monitoring, patient safety, and cancer care, to name a few.

Challenges for Future Development

There are many challenges as we go forward, including developing ways to collect data that may be important such as risk factors in chronic disease that may not be normally recorded in EMRs. Examples of this are ethnicity, occupation, education, and income. Recording these risk factors in the EMR where they are easily extractable (rather than as part of the encounter text) is a future goal. A solution to this may be to develop templates that would be acceptable to the physician and be sufficient as an encounter note but would organize data for more easy retrieval from the EMR. Templates are on our list for future development.

As a surveillance system for chronic disease in primary care across Canada, the data quality is a prime consideration. We are developing processes for assessing data quality and will need constant vigilance. Other ongoing issues include refining our approach to estimating practice denominators; the establishment of representativeness within the patient, practice, and regional network populations with whom we collaborate; and to consider broadening our vision of primary care to other health professionals in independent practice.

The growth of EMR use in family medicine in Canada (22% of practices in 2004²³) provides an important opportunity to collect more accurate, complete, and timely data than traditional billing-based surveillance systems without significantly increasing sentinel physician workloads. Given the resources and will of governments, the number of physicians using EMRs will increase during the next few years. We are also considering how many sentinel practices we will need to be able to provide reliable national and provincial estimates of chronic disease. For this we need to have networks in every province or region and sentinels in the 3 northern territories. As well rural and urban sentinels, representing provincial demographics for both physicians and patients will be needed to provide generalizable estimates. We do not expect this to be attainable in the short term, and statistical standardization techniques will be required. Finally, as we develop a larger database and gain more experience with the data, collaborations with networks in other countries will bring potential for international comparisons.