Sentinel Patients: Value of an Information System Collecting Patient's Own Report on His Healthcare Management

Main lines of the study. Evolution of patient's health after hospital discharge is poorly documented. We hypothesize that information of good quality on this concern might be accessible at a relatively low cost from patients, further referred to as sentinel patients, who would voluntary report their health evolution on a dedicated website.

The SENTIPAT study presents as a proof of concept study. It will collect patient's health evolution during a post-hospital discharge period of 45 days. The study presents as a randomized trial comparing patients' reports issued from two different methods in patients returning home after hospital discharge and having at disposal an internet connection available at home (further referred to as "eligible patients"):

in the " Internet " arm, sentinel patients will be solicited to directly report their data on a web-site especially devoted for collecting these data while in the " Telephone " arm, the same data will be collected through a telephone interview 45 days after hospital discharge. Investigations will concern appearance/worsening of symptoms, healthcare resources utilization, quality of life, and patients' satisfaction.

Adverse events. An adverse event is defined as non-related to patient's disease but associated with a default or excess of care. A clinically significant event is defined as one of the following events: new symptom(s) appearing after hospital discharge or worsening for symptoms present at hospital discharge, unscheduled rehospitalization or consultation whatever the reason for it, death (patients lost to follow-up will be investigated for their vital status). A validated event is an event considered as adverse by a validation medical committee, blind in terms of the source declaration whatever the source (sentinel patients or telephone survey). In addition, the committee will classify each validated event according to a criterion of avoidability.

Design. The main hypothesis of the study is that the risk of clinically significant validated adverse event is at least as well reported by volunteer patients (sentinel patients: Internet arm) as by a specific cohort follow-up (Telephone survey arm). SENTIPAT presents as a multi-center (five departments within the same hospital) non-inferiority randomized trial comparing the internet and telephone arms. Enrollment of eligible patients will be performed on the basis of consecutive such patients being discharged from hospital in the participating departments. Internet/Telephone randomization of eligible patients is stratified per department.

In addition, in order to estimate the distance of representativeness of eligible patients, a substantial number of patients having no internet access at their home (further referred to as "non eligible patients") will also be recruited and compose a third arm where patients are not randomized: all non eligible patients will be investigated via a telephone interview 45 days after discharge, just as the patients in the Telephone arm. No intermediate analysis was planned.

Primary objective. The primary outcome criterion is the proportion of patients reporting at least a "clinically significant validated" adverse event (see definition below) occurring during the period of 45 days after hospital discharge. The observed proportions in the Internet and Telephone arms will be compared.

Secondary objectives.

The secondary objectives include:

1. related to the primary criterion:

• Nature, chronology, severity and avoidability of the clinically significant adverse events
• Proportion of validated adverse events issued from the Internet and Telephone reports
• Association between adverse event occurrence and a) patient's perceived quality of care/health (SF-36 questionnaire, perceived health state-visual analogic scale-at admission, discharge, and 45 days after discharge), and b) post-discharge continuity of care (information transmitted to the post-discharge referent physician(s) of the patient).

1. related to the proof of concept for sentinel patients:

• Proportion of eligible patients, active sentinels, their characteristics as compared to the patient population hospitalized in the departments participating in the study.
• Rate for internet equipment and utilization.
• Rate of participation and completion of the questionnaires.
• Estimate the costs of the sentinel patients' information system and compare these costs to those of a specific cohort survey.

1. related to the objective of assessing the representativeness distance of eligible patients:

• Comparing the whole set of variables at baseline (i.e. inclusion, day of hospital discharge) between non eligible and eligible patients (e.g., socio-demographic characteristics-age, sex, social support-, hospital stay characteristics-reason, length, medication prescribed at discharge-).
• Comparing the rate of adverse events avoidable or not between the three arms (Non Eligible, Eligible-Internet, Eligible-Telephone).
• Making the above comparison in relation with hospital stay features, post-discharge other features (i.e. excluding post-discharge adverse events).
• Measuring patient's satisfaction (SAPHORA-MCO© questionnaire) and quality of life. The observed measures in the three arms will be compared.
• Description of the healthcare resources utilized after hospital discharge (consultations including their delay).

Sample size. Calculation was based on non-inferiority between two collection sources (Internet vs Telephone) of clinically significant adverse events in eligible patients. According to Forster et al (Ann Intern Med 2003, Cmaj 2004) we retained the target of 20% patients presenting at least such an event, and an equivalence limit of 5%. A 80% power is obtained through the inclusion of 2050 patients, 1025 in the Internet arm, and 1025 in the Telephone arm. In parallel, we planned to enroll 500 non eligible patients. The protocol plans the total enrollment of 2550 patients, with 205, 205, and 100 patients in the Internet, Telephone and Non Eligible arms, respectively, in each of the five participating departments.

Statistical methods. The results of the study will be presented according to CONSORT recommendations adapted to non-inferiority trials. Patients of the Internet and Telephone arms will constitute the sample for all comparisons between internet and telephone, which constitute the primary objective of the study. This same population will be considered for the comparative cost study. Eligible vs non eligible patients will constitute the sample for studying the representativeness of eligible patients to be sentinel vs non eligible patients. Primary analysis will be conducted using an "intention to watch over" strategy, applying the rule "lost to follow-up implies absence of event in the corresponding patients". Such an imputation of missing values is the most appropriate for rejecting the non-inferiority hypothesis in case of an important imbalance in the rate of lost to follow-up between the Internet and the Telephone groups. However, intention to treat is not usual in non-inferiority trials. Therefore sensitivity analyses restricted to patients with follow-up data will also be conducted.

Analysis of the primary outcome will be made by determining the bilateral 95% confidence interval of the difference in the proportion of patients with at least a validated clinically significant adverse event between the Internet and the Telephone groups, and by excluding the value of 5% of this interval.

Considering secondary analyses, within each of the participating departments, between group comparisons will be made using a Mann-Whitney or a Kruskall-Wallis test for quantitative variables, using Fisher's exact test for categorical variables, and using the Log-Rank test for duration variables. Considering the whole departments, The Friedman test will be used for quantitative variables and the Cochran-Mantel-Haenszel test will be used for categorical variables. Generalized linear regression models will be used for multivariate analyses of the criteria within the participating departments (multiple linear modeling, logistic regression, survival proportional hazards model) and generalized estimating equations modeling (with exchangeable correlations) will be used on departments' pooled data. Analyses will be conducted using SAS and R software.

The primary perspective retained for the cost analysis is the hospital perspective. A more complete analysis using a societal perspective might be considered later, including opportunity costs supported by sentinel patients. Most cost values will be directly issued from the study and all those involving horary tasks will be valued according to the human capital approach. The costs of the sentinel surveillance will include the costs associated to the development and the maintenance of the associated information system. The costs of the telephone survey will include those related to call management organization and to operator human time. Time spent for patient selection, inclusion, collection of non-opposition agreement, event coding, validation, and data analysis will be measured in the two groups of patients. Cost data will be balanced with participation rates and with questionnaires' completion and quality rates. In the end, costs for the Internet and Telephone strategies will be contrasted with the corresponding detection rates of adverse events. The incremental cost-effectiveness ratio of the Telephone strategy vs a patient's sentinel website will be estimated. Two situations will be considered: either limited to the study, the website being non-perennial, or considering a more intensive use: for example, models will explore wider multi-center uses of the website and longer periods than that that of the study (the latters implying an increase in maintenance costs but also a return of investment for the website development). The cost-effectiveness ratios in the two situations may substantially differ in the two situations.