Variation in recruitment across sites in a consent-based clinical data registry: lessons from the Canadian Stroke Network

Reasons patients were not accrued into the Registry (Figure 1)

Overall, 70% of potential participants were approached and 72% of these were enrolled in the Registry, with an overall accrual rate of 50.5%. Logistical challenges in approaching patients accounted for 60% of non-accrual. Major reasons included: death before the patient could be approached (10%); discharge from hospital before being approached (10%); and inability to make contact with the patient or surrogate after more than 3 attempts (20%).

Of the patients approached, approximately 17% refused to participate. This is approximately 2.5 times greater than that encountered in pilot work (unpublished).

Eight percent of patients were unable to consent due to communication difficulties with no surrogate available. Another 3% of those approached were not administered the consent form because their mother tongue (and that of their surrogate) was not English or French.

Variation in approach, agree, and overall participation rates (Figure 2)

The percent of eligible patients approached varied across sites from 41% to 86%, with a mean of 70%. The agree rate averaged 74% of those approached (lowest 25%, highest 95%). Approach and consent rates were not correlated across sites.

Supports and constraints on data collection (Table 1)

There was considerable variation across sites in supports for and constraints on approaching patients. Only five sites (25%) had active support from the emergency department (e.g. notification of new patients, providing brochures for patients not admitted to hospital). These sites achieved a substantially higher approach rate than did the sites without such support (average 80.5% vs. 67%; p < 0.02).

We found approach rates to be lower:

• When lists of potential stroke patients could not be obtained from the emergency department or the wards (60.6% vs. 70.4%);

• When coordinators could not approach patients directly. In these cases, the physician responsible for care had to first approach the patient. (59.9% vs. 71.1%); and

• When coordinators could not make contact with the patients after they had left the hospital (65.7% vs. 72.0%).

None of these results was statistically significant, although this may have been due to inadequate statistical power. Approach rates were no different in sites where the local principal investigator actively participated in the recruitment and where the nurse recruiter worked closely with the stroke team.

Regulatory and governance context

Canadian federal and provincial data protection laws allow for exemptions from consent for research purposes where, among other conditions: (a) the research cannot be achieved without using personal information, and (b) it is impracticable to obtain consent [7]. However, none of the provinces' legislation provides clear guidance as to the circumstances under which obtaining consent may be deemed impracticable. In Alberta, Saskatchewan, and Ontario, legislation specifically identifies this to be the purview of the research ethics board (REBs) [8].

Article 3.4 of the Tri-Council Policy Statement (TCPS) – the document that articulates the standards in Canada for REBs of institutions receiving funding from any of the three major federal granting councils – states that the "REB may[our emphasis] also require that a researcher's access to secondary use of data involving identifying information be dependent on: (a) the informed consent of those who contributed data or of authorized third parties;..." [9]. No specific guidance is provided as to the criteria for determining whether or not consent should be required for secondary use of existing personal information for research. However, two of the fundamental guiding ethical principles articulated in the TCPS are respect for free and informed consent and respect for privacy and confidentiality. Therefore, to be consistent with the values, purpose, and protections advanced in the TCPS, the onus for demonstrating a reasonable exception to the requirement for consent should fall on the researcher.

In 2005, the Canadian Institutes of Health Research (CIHR) issued its Best Practices for Protecting Privacy in Health Research[8]. Element 3 of the document includes detailed guidance as to the factors to consider when determining whether or not a research project should receive exemption from consent for secondary use of personal information. One of the provisions is quite broad – if, due to the size of the population, the proportion likely to have relocated or died, or lack of continuing relationship with the data holder:

"...there is a risk of introducing bias into the research because of the loss of data from segments of the population that cannot be contacted to seek their consent, thereby affecting the validity of results and/or defeating the purpose of the study."

We recognize the risk that this provision runs the risk of becoming a "trump card" over privacy concerns. Accordingly, it places a heavy onus on REBs to determine when to allow use of the data without consent and when to press for increased quality control in recruitment. This is relatively simple when it has been demonstrated that the vast majority of the potential research participants would be willing to allow their information to be used. It is much more difficult when, as we found, a non-trivial proportion of people approached refuses to participate.

Section 39 of the 2004 Ontario Personal Health Information Protection Act permits the disclosure of personal health information without consent to "prescribed registries" for the purpose of statistical analysis [12]. A handful of registries, including the Registry of the Canadian Stroke Network, are among the prescribed registries [10].

Experience elsewhere

The limited published literature on recruitment suggests that challenges in variation in recruitment faced by the Registry of the CSN are not unique. While researchers associated with the Mayo Clinic in Minnesota were able to achieve consent rates in excess of 95% to participate in a broad cross-section of disease-registries, there was variation both across sites and by diagnosis [11, 12].

It appears, though, that consent-seeking alone is not exclusively responsible for incomplete accrual. Across 91 U.K. clinical databases listed under the Directory of Clinical Databases (DoCDat), completeness of patient recruitment appears to be similar for databases that do and do not require individual consent for enrolment [13].

In a different context, Gross and colleagues examined patient accrual for 172 clinical trials in four high-impact medical journals. They found very poor reporting of the patient accrual process, with only 31 studies (18%) screening from a consecutive series of patients [14].

Lessons learned

Many researchers will still need to obtain informed consent for patient participation in their registry projects – for example, where there will be direct patient contact, where genetic information will be included or linked, or in particularly stigmatizing medical conditions. Several lessons can be learned from our experience with developing a consent-based registry. These lessons are derived from the data presented in this paper and from our discussions with site coordinators and co-investigators:

(1) The consent process needs to be thoroughly pilot tested under 'real-world conditions' with gradual roll-out to participating sites. One should anticipate ample lead time to develop, test, and implement the entire concept – particularly the consenting process and staff responsibilities.

(2) Close communications need to be established early and maintained with research ethics boards and health care institutions. This is probably best accomplished through a single contact-person working with each REB and hospital from the outset of the project.

(3) Accountability requirements for those responsible for obtaining consent should be as consistent as possible. Nurse coordinators in this study had a dual accountability: to the central coordinator and to local site principal investigators.

(4) Consider staging the implementation process, so as to build on the successes of the less complicated recruitment scenarios. For example, from the outset, we tried to recruit patients with transient ischemic attacks. This was ambitious, as these particular patients usually were not admitted to the hospital, and they constituted a large proportion of our patients not approached.

(5) Use a multi-pronged strategy for recruitment when potential registry participants have multiple points of access or care trajectories (e.g. both inpatient and outpatient treatment). Obtaining consent may be more feasible when repeated outpatient visits allow increased contact and trust.

(6) Obtain firm support of those departments that have first contact with target patients (e.g. Emergency) to identify potential participants and provide them with information and support to implement screening processes.

(7) Consider random sampling strategies to reduce workload, rather than including all consecutive patients. This was a strategy we implemented in Phase 2 in institutions with particularly high volumes of stroke patients. We found this increased the approach rate.

(8) Ongoing monitoring and feedback on accrual help to increase and sustain higher accrual rates and interest.

(9) Consent forms in other languages and access to translators may be required for projects operating in jurisdictions with multi-ethnic populations. Usually, such hospitals have a roster of translators for such situations.

(10) Elicit ongoing patient feedback – particularly from those who hesitate or refuse to participate – to ascertain what concerns they may have. While some refusers may not wish to share this information, if this is done in a way that does not pressure patients, then it can provide valuable feedback.

Long-run changes are needed

Concern has been expressed elsewhere that, for multi-centered studies, the process of research ethics approval is very time consuming, with considerable duplication of effort and local idiosyncratic restrictions that offer little perceived gain [15–17]. In some countries, a centralized review process has been implemented for such multi-centered studies. While intended to streamline the review process, in some cases this has simply added another level of bureaucracy [18]. Even greater standardization of the process would be helpful. Assimilation of the CIHR privacy guidelines into the review process could help to harmonize the interpretation of acceptable recruitment practice.

In our study, the chief source of variation in administrative requirements came not from the REBs but from the data stewards – acute care hospitals. In particular, we found major differences in (a) ability to coordinate with the emergency department in the recruitment process; and (b) hospital policy as to whether, and at what point, potential registry participants could be approached to participate. In part, this can be resolved through education of health care institutions as to what is permitted by the law.

Looking forward, numerous registry and linked electronic health record data collection activities are being planned in North America and Europe. Perhaps it is time to re-think how we go about recruiting patients into these registries [19]. For example, would it be more efficient to shift responsibility for patient accrual from managers of individual projects to the institution and for a network of institutions to develop common protocols across institutions for obtaining consent for use of clinical records, obtaining biological samples, follow-up surveys, and linkage of clinical data with administrative records?