Graphical summaries

Summary statistics, however calculated, are unlikely to encapsulate all the subtleties of QL data. Therefore it is important to first examine the data graphically before performing detailed analysis. This will also help readers interpret the data. For example, plotting the scores against the time from randomization will give a better feel for the range of data and the variability of completion over time.

Figure 9.6 shows a scatter plot of data of HADS depression scores [21], calculated as the sum of seven individual item responses for each patient providing data during the first year from an MRC Lung Cancer Working Party trial, plotted against the actual assessment date. The scheduled assessment times are indicated for the first year as vertical hatched lines. The horizontal lines represent agreed definitions of'clinically normal' with a score of <7, those with 'clinical depression' with a score of >11 who would probably need

HADS score 20

Case

Borderline

i été* ¡** ! ^ ■ 1111111** *** «**«*# | * •frtfr--^- II [ ! 11 I 1 '[ [ § )(

m • ***** * j #*j * j ** * j* * "MiHmiHH rtr"i!~tfrl' tii^ i* j %

0 100 200 300 400 Days from randomization

Fig. 9.6 Scatter plot of HADS depression score against actual assessment by the patient for the ten scheduled assessments in the first year (Reprinted with permission from John Wiley & Sons Ltd from reference [21]).

medical assistance whereas the 'borderline' patients with HADS of 8, 9 or 10 may require further assessment. This plot illustrates the increasing amount ofmissing data over time (after the baseline assessment) and the departure ofassessments from the planned times.

It can also be useful to plot individual patient profiles as this will indicate the patterns of change and the number of patients with incomplete data.

Figure 9.7 shows the data for the symptom 'lack of appetite' for twelve patients in the CHART lung cancer trial [22], six from the conventional radiotherapy group and six from the CHART group. These data shows the variability both within and across patients in the profile of 'lack of appetite' over time and also show that missing data are common. This is useful to remember, because summary data averaged over patients is unlikely to show such variability. The figures in the right-hand corner of the boxes are explained below.

It is possible to plot the percentage of patients falling into a particular category (or categories) - i.e. a dichotomy of the scale, to show the proportion of patients experiencing a particular symptom over time. Alternatively we could use another statistic, such as the mean score. However, such plots, which are often found in the literature, can be very misleading as the number of patients completing forms at each timepoint will differ and usually decrease over time, and it is important to indicate the number of patients returning forms at each time on the plot. The obvious way around this is to restrict the dataset to only those patients who have returned complete data up to a certain timepoint. However, the sample size may become greatly reduced, and as these data will have come from patients who survive and complete forms, and thus may have better physical and psychological status, the result may not be generalizable to the whole trial population.

CHART Patient A

CHART Patient B

CHART Patient C

CHART Patient D

CHART Patient E

CHART Patient F

Conventional Patient A

Conventional Patient B

Conventional Patient C

Conventional Patient D

Conventional Patient E

Conventional Patient F

Conventional Patient E

Score

Fig. 9.7 Lack of appetite profiles for a sample of six patients from each treatment group (for the CHART trial) - A point represents an observed score and the lines between points are interpolations of the score between times.

Score

Score

Score

Score

Score

Score

Score

Score

Score

Score

Score

Bias will occur if more patients drop out of one arm. Nevertheless, such plots (even if just produced for internal consideration) can provide valuable insight into the data.

Graphical summaries do not provide statistical tests of hypotheses, nor do they provide summary estimates. Nevertheless, because exploratory and descriptive analyses are often less concerned with significance testing, graphical methods may be especially suitable for presenting QL results. These have a number of advantages over purely numerical techniques. In particular, judicious use of graphics can succinctly summarize complex data that would otherwise require extensive tabulations and can clarify and display the complex interrelationships of QL data. At the same time graphics can be used to emphasise the high degree of variability in QL data. This contrasts with numerical methods, which may often lead to results being presented in a format which leads readers to assume there is greater precision than the measurements warrant.

Cigarette Crusher

Cigarette Crusher

Get All The Support And Guidance You Need To Permanently STOP Being A Slave To Nicotine And Cigarettes. This Book Is One Of The Most Valuable Resources In The World When It Comes To Easy Ways To Eliminate Smoking Addiction And Revitalize Your Body.

Get My Free Ebook


Post a comment