Historical controls

Here, a new treatment is tried on a group of patients and results are compared with a 'similar' group of patients previously treated on a standard therapy. Previously treated patients might include

♦ literature controls - a comparison with published results of similar trials

♦ consecutive groups of patients treated within the same hospital over different time periods

There are problems with both approaches, first and foremost that of patient selection. Early trials may well assess the impact of treatment in patients with very advanced disease, but will eventually consider those patients (possibly with much less advanced disease) for whom the treatment is actually intended. It is understandable that such early studies will focus on those patients who, through disease characteristics and general fitness, would be expected to respond most positively. Identifying a control group in whom the same selection criteria have been applied is more difficult. A well established treatment is likely to have been applied to a much wider variety of patients in more recent years.

Even if patients are unselected, other factors may have changed over time (for example, supportive care or antibiotic use). The criteria for assessing 'response,' including staging procedures, may have changed over time - or data on the historical group of patients may not have been recorded in sufficient detail for comparison with current patients to be made. The type of patient being seen may have changed. This might occur through specific changes in referral practice, for example specialist centres may be referred more 'difficult' cases. Perhaps even more difficult to account for are changes in the overall pattern of disease stage, for example the shift towards presentation at an earlier stage resulting from screening for breast, prostate or bowel cancer.

Two examples follow, the first illustrating a fairly typical example of the impact of patient selection, the second a more subtle issue.

Example 1: Treatment of malignant brain tumours. In the late 1980s, the MRC was considering a new trial in high-grade glioma, which would address whether or not some form of chemotherapy could improve survival when given in addition to the standard treatment of surgery followed by radiotherapy. One of the options was a new drug which had undergone phase II testing in one UK hospital. There was a considerable amount of data, both from previous MRC trials and other sources, establishing the 2-year survival rate on standard therapy at no more than 10-15 per cent. The new drug study showed a 2-year survival rate of approximately 35 per cent. Clearly this was very encouraging. However, the previous trials had not only provided data on the expected survival on standard therapy (Fig. 3.1a), they had also provided data on prognostic factors. In particular, data from two previous MRC trials had been used to derive a prognostic index [9] which was able to classify patients into six groups with varying probabilities of survival (Fig. 3.1b).

When the patients on the new drug were assessed and categorized into the prognostic groups defined by this index it was clear that they fell into the best prognostic groups, and that their survival did not differ substantially from that expected in patients with similar characteristics but undergoing conventional treatment. While this alone did not

6 12 18 Months from start of radiotherapy

(b)

1.0

0.9

0.8

0.7

O)

a

0.6

>

0.5

>

ID

0.4

Ln

0.3

0.2

0.1

0.0

Fig. 3.1 (a) Overall survival of patients with malignant glioma undergoing radical radiotherapy. (b) Survival according to prognostic group.

■ Score

0-5

Score

6-10

- Score

11-15

Score

16-21

Score

22-30

Score

>30

6 12 18 Months from start of radiotherapy

Fig. 3.1 (a) Overall survival of patients with malignant glioma undergoing radical radiotherapy. (b) Survival according to prognostic group.

eliminate the new drug as a potential component of the new trial, it did identify the need for cautious interpretation of the pilot study data.

Such situations are common, although good data on baseline survival and on prognostic factors against which new data can be assessed are unfortunately not always available. While this situation is one in which the potential biases involved in historically controlled studies are fairly easy to anticipate, the issues are not always so clear.

Example 2: Testicular cancer. The outcome of treatment for metastatic testicular germ cell tumours improved substantially during the 1970s through the introduction of the drug cisplatin. During the 1980s combination chemotherapy regimens were developed and these too, possibly combined with increased experience in treating the disease, led to improvements in outcome. More recently, with standard treatment established as the BEP (bleomycin, etoposide, cisplatin) regimen, there have been no clear improvements in outcome. Meanwhile, post-operative surveillance had become a widely accepted form of management for early stage disease. In such a situation, one might consider the following proposal:

♦ Between 1990 and 1993, every patient with stage I disease (disease confined to the testis) at a particular hospital was treated by surgery alone.

♦ The same hospital treats the next 100 stage I patients with surgery followed by BEP chemotherapy.

This is a rare disease in which randomized trials might be thought difficult. Some of the usual concerns with the use of historical controls appear to be absent - for example, overall survival has not changed recently, the type of primary treatment (surgery) has remained the same, and the endpoint - survival - is clear and objective. It might therefore seem reasonable to assume that, if survival improves, it must be due to the treatment, because the patients are unselected, and no other aspect of treatment has changed.

However, there is possibly a more subtle bias here. Patients undergo radiological procedures to establish their disease stage. In recent years, this would include chest CT to look for lung metastases, whereas in the past this would more often have been done using chest X-rays. If chest CTs are able to identify patients with lung metastases too small to be identified using chest X-rays, then a group of patients who would previously have been classified as stage I, but who would have developed obvious lung metastases in due course, would in the present day be reclassified as having advanced disease immediately. The prognosis of the present-day group of stage I patients is better than that of'historical' stage I patients because a high-risk group is now excluded. However, the prognosis of the present day advanced disease group may also have improved, because the additional patients now falling into this group have minimal disease. It is possible in this situation for survival in both subgroups (stage I and advanced disease) to improve, while overall survival stays the same. Although this seems counter-intuitive at first, we are simply shifting patients from one group to another. A hypothetical numerical example will illustrate this.

Box 3.4 shows survival rates by disease stage for 1000 patients treated in the 1980s and another 1000 patients treated in the 1990s.

We see that in the 1990s, fewer patients are classified as stage I, but those that are have a slightly higher survival rate (99 per cent) than those diagnosed as stage I in the 1980s (95 per cent).

0 0

Post a comment