Rapid Responses to:
|
|
Rapid Responses: Rapid Responses published:
-
![[Read Rapid Response]](/icons/misc/sectionDOWN.gif)
Extrapolation of clinical trials is not as straightforward as it might seem at first glance
- Zvonko Rumboldt, Vedran Carevic (21 May 2005)
-
Confidence intervals not correctly reported
- Claudia Schmoor, Manfred Olschewski, Martin Schumacher (29 May 2005)
-
Confidence intervals for adjusted results
- Gunn Elisabeth Vist, Kåre Birger Hagen, PJ Devereaux, Dianne Bryant, Doris Tove Kristoffersen, and Andrew David Oxman (31 May 2005)
|
|
|||
|
Zvonko Rumboldt, Professor of Medicine and Clinical Pharmacology Split University School of Medicine, 21000 Split, Croatia, Vedran Carevic
Send response to journal:
|
Sir, The paper „Systematic review to determine whether participation in a trial influences outcome“, particularly following the relevant report of the GRACE investigators and the timely editorial by professor Green prompted us to this comment. The common message is that clinical trials are highly predictable for clinical practice, but only if the latter patients are strictly comparable to the former. Let us illustrate this message with a recent example from Croatia, concerning acute myocardial infarction with ST elevation (STEMI). During a 6-year period (January 1, 1996-December 31, 2001) 1,608 patients were admitted to Coronary Care Unit, Department of Medicine, Split University Hospital, Croatia, with STEMI; 284 (17.7%) of them died during the hospital stay. Anterior wall STEMI was responsible for the majority of fatal outcomes (134/284 or 47.2%). Death rate was higher among the elderly (greater than 70 years of age), accounting for 197/284 (67.6%) deceases. The hospital mortality was unexpectedly, albeit marginally higher among the thrombolysed patients (streptokinase 1.5 MU/60 mins; 21.5% vs. 17.1%; P=0.231). There was no difference in streptokinase administration between the patients below and above 70 years of age (108/798 vs. 111/810; P=0.921) but there was a marked discrepancy in the way of dying relative to thrombolysis; free wall cardiac rupture with electromechanical dissociation was much more prevalent among the thrombolysed patients, while cardiogenic shock predominated among patients who were not thrombolysed (P<0.001). There were 67/284 or 23.6% of cardiac rupture deaths overall; it was more prevalent among the thrombolysed subjects (20/47 or 43.6%), and exceedingly high if they were above 70 years of age (13/20 or 65.0%; P<0.0001). Free wall rupture was diagnosed clinically and documented with echocardiography; in 7 of them it was confirmed at autopsy as well. A significant difference in time to cardiac rupture, relative to streptokinase administration was observed as well: in thrombolysed patients it appeared much earlier (P<0.001).Two of the SK treated patients (2/219 or 0.91%) developed radiologically confirmed intracranial haemorrhage; one of them died, autopsy was not done. Since prompt angioplasty is simply beyond the capacities of transitional/developing countries, such as Croatia is, urgent thrombolysis is essential. Surprisingly, we have observed no significant difference in mortality between so treated and untreated patients. The reason may be due to the unacceptably long pain-to-needle time, which in our hospital averages some 4 hours, while optimal results are achieved within the first 60-90 minutes, emphasizing the role of rapid, even extramural intervention. Although reperfusion substantially improves cardiac function and prognosis in STEMI, our results suggest that in addition to the risk of hemorrhagic stroke it increases the probability of early cardiac rupture, particularly if instituted in the elderly, after the first hour or two have elapsed from the onset of heart attack. Let us conclude: 1) In transitional and developing countries, prompt primary percutaneous coronary intervention is out of reach for the vast majority of AMI patients. 2) Coronary revascularization must therefore be achieved with urgent administration of a thrombolytic agent (mostly streptokinase): the importance of early diagnosis and prompt institution of such therapy, even before hospital admission (particularly if transportation to the nearest hospital is cumbersome) can not be overstressed. 3) Any delay markedly decreases the effectiveness of thrombolytic intervention and increases the probability of fatal complications (cardiac rupture, intracerebral haemorrhage), particularly among the elderly, making sometimes treatment worse than the disease. 4) Extrapolation of clinical trials to clinical practice is not as straightforward as it might seem at first glance. References: 1. Vist GE, Hagen KB, Devereaux PJ, Bryant D, Kristoffersen DT, Oxman AD. Systematic review to determine whether participation in a trial influences outcome. BMJ 2005;330:1175-81. 2. Green L. Benefits of early invasive treatment for acute coronary syndromes: lost in transition? BMJ 2005;330:E351-2. 3. Novak K, Polic S, Rumboldt M, Lukin A, Carevic V, Rumboldt Z. The mechanisms of death in elderly patients with acute myocardial infarction exposed to fibrinolytic therapy. In submission. 4. Poliæ S, Perkoviæ D, Štula I, Punda A, Lukin A, Rumboldt Z. Early cardiac rupture following streptokinase in patients with acute myocardial infarction: retrospective cohort study. Croat Med J 2000;41:303-5. Competing interests: None declared |
|||
|
|
|||
|
Claudia Schmoor, Statistician Centre of Clinical Trials, University Hospital Freiburg, 79110 Freiburg, Germany, Manfred Olschewski, Martin Schumacher
Send response to journal:
|
The confidence intervals reported in figures 2 and 4 of Vist et al. [1] for our results (Schmoor et al. [2]) are far too narrow. The correct confidence intervals as they were sent to the authors are as follows. Figure 2: Relative Risk (95% CI) Schmoor 1996a: 1.03 (0.69 to 1.52), Schmoor 1996b: 1.01 (0.72 to 1.43), Schmoor 1996c: 0.72 (0.42 to 1.23), Schmoor 1996d: 0.94 (0.50 to 1.74), Schmoor 1996e: 1.20 (0.78 to 1.83), Schmoor 1996f: 1.00 (0.56 to 1.78) Figure 4: Relative Risk (95% CI) Schmoor 1996a: 1.00 (0.64 to 1.57), Schmoor 1996b: 1.18 (0.82 to 1.72), Schmoor 1996c: 0.88 (0.49 to 1.58), Schmoor 1996d: 1.54 (0.85 to 2.80), Schmoor 1996e: 1.29 (0.83 to 2.01), Schmoor 1996f: 1.10 (0.62 to 1.97) In contrast to the results presented by Vist et al. [1], all these confidence intervals include the value one, indicating non-significant differences between the compared patient groups. This result could be expected, because the analyses were adjusted for prognostic factors. References: 1. Vist GE, Hagen KB, Devereaux PJ, Bryant D, Kristoffersen DT, Oxman AD. Systematic review to determine whether participation in a trial influences outcome, BMJ 2005; 330: 1175-1181. 2. Schmoor C, Olschewski M, Schumacher M. Randomized and non- randomized patients in clinical trials: experiences with comprehensive cohort studies. Stat Med 1996; 15: 263-71. Competing interests: None declared |
|||
|
|
|||
|
Gunn Elisabeth Vist, senior researcher Norwegian Health Services Research Centre, PO Box 7004, 0130 Oslo, Norway, Kåre Birger Hagen, PJ Devereaux, Dianne Bryant, Doris Tove Kristoffersen, and Andrew David Oxman
Send response to journal:
|
We thank Claudia Schmoor for noticing the error that occurred in our calculations of confidence intervals (CI) of the adjusted results in our review. This error resulted in the CIs of the adjusted results being erroneously narrow. The correct CIs of the adjusted results are as follows: Figure 2, adjusted results for main outcome Adjusted main results RR (Confidence Interval) Davis 1985 0.39 (0.18 to 0.83), Feit 2000a 1.17 (0.96 to 1.42), Feit 2000b 0.94 (0.74 to 1.20), Mosekilde 2000a 1.15 (0.60 to 2.19), Mosekilde 2000b 1.32 (0.37 to 4.72), Schmoor 1996a 1.03 (0.69 to 1.52), Schmoor 1996b 1.01 (0.72 to 1.43), Schmoor 1996c 0.72 (0.42 to 1.23), Schmoor 1996d 0.94 (0.50 to 1.74), Schmoor 1996e 1.20 (0.78 to 1.83), Schmoor 1996f 1.00 (0.56 to 1.78), Williford 1993 0.60 (0.42 to 0.86). Figure 4, adjusted results for mortality Adjusted mortality RR (Confidence Interval) Davis 1985 0.39 (0.18 to 0.83), Feit 2000a 1.17 (0.96 to 1.42), Feit 2000b 0.94 (0.74 to 1.20), Schmoor 1996am 1.00 (0.64 to 1.57), Schmoor 1996bm 1.18 (0.82 to 1.72), Schmoor 1996cm 0.88 (0.49 to 1.58), Schmoor 1996dm 1.54 (0.85 to 2.80), Schmoor 1996em 1.29 (0.83 to 2.01), Schmoor 1996fm 1.10 (0.62 to 1.97). In comparison to previously presented results, the CIs for four more of the main outcomes cross one (Feit 2000a, Feit 2000b, Schmoor 1996c and Schmoor 1996e). Thus, of the 73 comparisons reporting dichotomous main outcome, 63 reported no significant difference between outcomes for patients treated in RCTs and those receiving similar treatment outside RCTs; eigth reported significantly better outcomes for patients treated in RCTs, and two reported significantly worse outcomes for patients treated in RCTs: rather than ten statistically significantly better and four worse as we reported. For mortality, five more CIs cross one (Feit 2000a, Feit 2000b, Schmoor 1996b, Schmoor 1996d and Schmoor 1996e) than we reported. Thus, of the 32 comparisons reporting mortality, 29 reported no significant difference in mortality for patients treated in RCTs and those receiving similar treatment outside RCTs; three comparisons reported significantly lower risk of mortality for patients treated in RCTs, and none reported significantly higher risk of mortality for patients treated in RCTs: rather than four statistically significantly better and four worse as we reported. These errors do not affect our main conclusion: participants in RCTs had similar outcomes to comparable patients who received the same or similar treatment outside the trial. We apologise for not having detected these errors prior to publishing our review. Competing interests: None declared |
|||