Influence of follow-up response on risk-factor analysis

Long-term analyses after coronary artery bypass grafting (CABG) are used to investigate therapeutical options and factors influencing the natural course of ischemic heart disease. In general, long-term studies require a follow-up. Dependent on the interval between the intervention and the follow-up procedure a certain amount of patients is lost to follow-up. The aim of the present study was to examine the influence of incomplete follow-up on conclusions regarding the postoperative patient outcome. For the investigation, the same statistical methods were applied to the data accumulated by the 70% and by the 90% responses.

2012 patients underwent isolated CABG between June 1988 and December 1992. For data acquisition, tools of the HVMD (Heidelberger Verein für multizentrische Datenanalyse e.V.) were used. Analyses were performed using tools of SAS (Statistical Analysis Systems, Inc.). The parametric, time-adjusted hazard function method was employed. A first follow-up questionnaire, was distributed six months after operation with a 97.8% response. In February 1997 the same questionnaire was sent to patients and their general practioners. The primary response to that was 68.9% (approximately 70%). Then another mailing of the same questionnaire and phone calls to patients and their home doctors raised the response to 93.7% (approximately 90%).

The mean follow-up was 1378 days in the group with 70% response and 1682 days in the group with 90% response. The parametric, time-adjusted hazard function showed a very similar pattern of factors in the early phase of both groups. In the 90% response group, the intermediate phase reached a higher relative influence than in the 70% response group. The relative influence of the late phase showed an inverse pattern. In the multivariate analysis most of the variables which had been identified by the 70% response model reappeared in the 90% response model. However, there were some potentially important and interesting differences.

The results indicate the necessity to carefully consider the acceptance of incomplete follow-up for differentiated risk adjustment.