Introduction
What is new?
- •
Fifty-nine percent of episodes of LFU were followed by a return to care, either at the same center or at a different center.
- •
Individuals returning for care were more likely to have started HAART, have higher CD4 counts and viral loads, and be younger. They were less likely to have had a previous LFU episode.
- •
Data linkage decreases the number of individuals potentially LFU and increases periods of follow-up.
Loss to follow-up (LFU) is a major concern in any clinical study, as it not only reduces the power of any statistical analyses but also has the potential to introduce bias if individuals' LFU differ in any respect from those remaining under care (often referred to as attrition bias). Thus, an essential part of any study write-up is the inclusion of the degree of LFU [1], [2], [3]. In any study where visits are determined at regular fixed intervals, such as a randomized trial or some epidemiological cohort studies, it is relatively easy to describe the proportion LFU at any point. However, this information may be more difficult to summarize in an observational database formed using data collected as part of routine clinical care. In such settings, the data collected often reflect assessments of patients at visits dictated by clinical need—these visit times may vary from patient to patient and may be infrequent. In such a setting, it may be difficult to differentiate patients who are genuinely LFU from those who are infrequent attenders. Currently, patients are often defined as LFU simply by selecting a time period, often arbitrary, over which they had not attended for care.
The UK Collaborative HIV Cohort (UK CHIC) Study is a large multicenter cohort comprising data from some of the largest human immunodeficiency virus (HIV) clinics across the United Kingdom. The main benefit of such a multicenter cohort is that it can provide information on long-term follow-up from a larger and more diverse group of patients than would be possible in a single-center cohort. The greater number of records result in increased statistical power for studies of less commonly occurring clinical events. However, some cohort participants are known to have attended more than one participating clinic. If data from these individuals remain unlinked, the results of analyses may be biased by the inclusion of multiple records relating to the same individual. Not only would this artificially increase the denominator for many analyses, it would also lead to the underestimation of risk estimates if multiple records for an individual overlapped in follow-up and if time periods were counted twice in the analyses.
In the UK CHIC Study, we use a linkage process that merges records thought to relate to the same person. As part of this project, we aimed to assess the validity of a commonly used definition of LFU by describing the number of patients potentially LFU who reentered the data set, either at the same center or at a different clinical center. We then identified risk factors for permanent LFU (compared with reentering the data set) and factors associated with reentering the database through the same clinic as opposed to a different clinic.