Features of QL data

Multi-dimensionality. All widely used questionnaires consist of multiple items, but for simplicity and robustness rather than analyse individual items, the common approach is to combine items into subscales. Scoring manuals to form these subscales are available for all major questionnaires, see, for example, Refs [15,16]. There is a psychometric reason for asking more than one question about one aspect of QL as discussed in Chapter 6, but there is also a statistical reason, in that error terms for individual questions tend to be large and summation across questions reduces the variability and therefore improves reliability.

Compliance and missing data It is important to present information on patient compliance and to assess whether the compliance is approximately of the same level and follows a similar pattern in each treatment group. It is also important to show how representative the population from whom QL data have been collected is of the whole trial population. Hopwood and colleagues [17] have defined compliance as the number of forms completed as a proportion of those expected, where the anticipated number of forms is the total number that should have been completed by each patient according to the protocol schedule but taking account of the date of death.

Not only is it important to indicate what forms have been received, but whether they were completed at the scheduled times. Let us assume, for example, that the scheduled QL assessment day is day forty-two. If the corresponding QL form is not completed until say, day seventy-two, the responses recorded may not reflect the patient's QL at the timepoint of interest. However, it is necessary to recognize the variation in individual patients' treatment and follow-up, and so a reasonable time frame (or window) should be allowed around each scheduled protocol assessment timepoint, as introduced in Chapter 6. Clearly such time windows may be different for different timepoints. For example for the pre-treatment quality of life form the window may be narrow, such as seven days, and asymmetrical, i.e. before the start of treatment. Alternatively, for a three-monthly form a symmetrical window of one month may be used, with possibly wider windows for later timepoints which are more widely spaced.

One of the major problems with quality of life data is missing data. There are two aspects of missing data: missing forms and missing items. In cancer trials, some missing forms are unavoidable. When such forms are termed to be 'missing completely at random' (MCAR), then the missing data do not depend on any characteristics of the patient or the condition of the patient, and the missing data can be relatively simply imputed. If the forms are 'missing at random' (MAR), that is not related to the condition of the patient but related to other observed factors e.g. performance status, it is possible to model what would have happened if those data had been collected, based on what is known about the patient and what is known about other patients. However, in the assessment of QL the proportion of missing forms increases as patients' health deteriorates and particularly as they approach death. Thus the majority of forms are 'not missing at random' (NMAR), which means the reasons for the missing data are related to the unobserved factors, for example the condition of the patient. The standard methods for imputing missing data, extrapolating in some way from within (or outwith) the data set, cannot be used. It would be illogical to impute data for patients who are ill and unable to complete forms from patients who are fit and well and have completed forms. There are some methods of multivariate analysis that allow missing values to be ignored, and some statistical methods of imputing NMAR data, but in QL analyses all such methods must be used with extreme caution.

As opposed to missing forms, very few papers report the extent of missing items on forms or how these were handled in the analysis. Trials groups such as the Medical Research Council and the European Organization for the Research and Treatment of Cancer regularly report between 0.5 per cent and 1 per cent missing items, which seems trivial. However, as most questionnaires include 30-40 items, this can mean that at each assessment between 15 and 40 per cent of forms have missing items, and if patients are expected to complete 4-6 questionnaires, there is the potential for there to be no patients at all with complete data at all timepoints.

Most standard questionnaire manuals suggest methods of dealing with missing items so that missing individual items do not necessarily mean the loss of data. For instance, if the patient has answered three of four items relating to, say, anxiety, in a positive way, it is reasonable to presume they would have answered the 4th in the same manner. The recommendation is to estimate summary scale scores using the mean of the other observed scale items. However, caution should be exercised in doing even this. Most of the questionnaire manuals [15,16] include an algorithm for allowing imputation. They usually state that as long as data are available for at least 50 per cent of the items the scale can be formed and standardized to 0-100 accordingly. Nevertheless, this has dangers, and Fayers et al. [18] summarized a number of checks that should be made (Box 9.6) before applying such imputation (Box 9.7).

10 Ways To Fight Off Cancer

10 Ways To Fight Off Cancer

Learning About 10 Ways Fight Off Cancer Can Have Amazing Benefits For Your Life The Best Tips On How To Keep This Killer At Bay Discovering that you or a loved one has cancer can be utterly terrifying. All the same, once you comprehend the causes of cancer and learn how to reverse those causes, you or your loved one may have more than a fighting chance of beating out cancer.

Get My Free Ebook


Post a comment