Among 28 acute care not-for-profit hospitals in Connecticut, we identified 12 with more than 300 beds. One was not invited to participate because, based on past experience, we believed it would decline. Two chose to participate as independent sites outside of the LION project. We added 2 hospitals with less than 300 beds: a rehabilitation hospital and an acute care hospital with strong records in stroke research. A total of 11 hospitals, therefore, were invited to participate and initially accepted.
An IRB application was ultimately filed at 10 of the 11 institutions. One of the 11 was excluded when its Research Review Committee objected to the LION concept and would not authorize subsequent review by its IRB. By June 30, 2007, 8 of the remaining 10 hospitals were active and 2 were not. One small hospital was terminated when active surveillance revealed few eligible patients. Another had not begun to identify potential IRIS participants. Essential features of the 10 institution are described in .
Table 1 General hospitals and rehabilitation hospitals to which an IRB application was submitted
The IRBs at all 10 institutions were asked to approve the surveillance system described above that would give personnel from the Network Coordinating Center access to logs of patients admitted or discharged with the diagnosis of stroke or TIA, their names, the names of their physician, and contact information. Approval was granted at 5 institutions and denied at 5. The 5 denying institutions approved strategies involving surveillance by local hospital staff and referral to the Network Coordinating Center. This strategy succeeded at 2 institutions and failed at 3. At the successful sites, surveillance was completed by discharge planners who identified eligible patients during their regular work (institution 7) or a research associate who worked for the facility but was paid separately for work on the IRIS trial (institution 10).
At the 3 other institutions, surveillance was attempted by diagnosis coders (institution 4), the principal neurologist (institution 6), or case mangers (institution 8) during their routine work activities. After data showed slow recruitment or incomplete surveillance, the IRBs at institutions 4 and 6 approved the originally requested surveillance system. The IRB at institution 8 approved a strategy similar to that for institution 10 in which a member of the hospital staff was paid to perform active surveillance for IRIS, but work had not started by June 30, 2007.
Thus, the preferred (template) protocol for patient identification by which patient lists were provided to researchers in the Network Coordinating Center was ultimately approved by 7 of the 10 hospitals participating in the LION project. The other 3 hospitals require employees of the institution to locally review admission or discharge logs and medical records to preselect eligible patients for referral to the Network Coordinating Center.
Preliminary eligibility was checked before direct patient contact for patients from most LION hospitals. Only the minimum information necessary was gathered. Strategies included remote review of electronic medical records by IRIS staff (3 institutions), on-site review of paper medical records (1 institution), and on-site review of medical records by in-house personnel (2 institutions).
Nine of the 10 hospitals approved the preferred (template) strategy for first and subsequent contact of potential participants by IRIS staff. Hospital 9 required us to send a letter to patients in advance of our telephone contact; patients could call to opt out of the telephone call.
Performance data for the Connecticut LION and for each of the 10 active, inactive, or terminated member institutions are reported in . A total of 4,708 patients were identified to the Network Coordinating Center, among whom 763 (16%) were determined to be eligible for a consent visit. Of the 763 patients, 708 (93%) agreed to an in-person consent visit, usually in their home. As a result of the consent visit, 441 of 708 (62%) were found to be eligible for enrollment. A high proportion of eligible patients consented to participate in the trial (281/441 = 64%).
Table 2 Performance of active member institutions in the Connecticut network during January 1, 2005, to June 30, 2007
Member institutions in the Connecticut LION were, on average, more productive in terms of randomizations than other institutions in the IRIS network. Among 104 institutions outside Connecticut that participated in the trial at any time before June 30, 2007, the average monthly randomization rate was 0.33 participants (range 0–1.88). Adjusted for bed size of enrolling institutions, the average monthly randomizations per 1,000 beds was 0.56 (range 0–4.53). Among the 75 institutions outside Connecticut that enrolled at least one patient, the average monthly randomization rate was 0.46 participants (range 0.03–1.88); average monthly randomizations per 1,000 beds was 0.76 (range 0.04–4.53). Among the nine active or terminated sites in the Connecticut LION, the average monthly randomization rate was 0.51 participants (range 0.13–1.21); average monthly randomizations per 1,000 beds was 1.52 (range 0.38–5.58). After adjusting for bed size, hospitals in Connecticut remained more productive compared with the 75 outside Connecticut that have enrolled at least one patient (1.52 randomizations/1,000 beds/month compared with 0.76 randomizations/1,000 beds/month, Wilcoxon rank-sum comparison p = 0.03). If the Connecticut LION is considered as one site, the average monthly enrollment is 4.0.
displays essential demographic and clinical characteristics of patients enrolled at the Connecticut LION compared with other sites. Patients from Connecticut were older, more disabled, and less likely to be black. Patients from Connecticut were also more likely to enter the trial with a stroke, rather than a TIA, although the difference was not significant.
Table 3 Baseline characteristics of patients from the Connecticut LION compared with other sites
Experience with home visits.
Among 708 patient visits performed by research nurses in the Connecticut LION, almost all took place at home. Nurses never found themselves in situations they perceived to be dangerous. One nurse sustained injuries in a car accident while returning from an assignment.
Key indicators of data quality for the Connecticut LION, compared with other IRIS sites, are shown in . Data quality was good at all sites, but trends favored the Connecticut LION, particularly for completion of annual phlebotomies and pill bottle returns. Returned bottles were used to calculate medication adherence. Not shown in is average time from stroke or TIA to randomization, which was 71 days for the Connecticut LION, compared with 86 days for other sites (p < 0.0001).
Table 4 Selected indicators of research and data quality
The average total cost to randomize and follow one participant from the start of funding July 1, 2004, through June 30, 2007, including facilities and administration costs, was $8,697 for the Connecticut network, compared with $7,198 for the other sites.
Among 4,708 patients identified and more than 2,000 direct contacts, we encountered only 2 patients or relatives who expressed discomfort with our approach. The son of 1 patient had privacy concerns that were quickly resolved when we explained the IRB approval process and our compliance with HIPAA. Another potential participant became upset after learning we could not pay for his participation. No complaints against the trial have been filed with member institutions or their IRBs.