To the Editor: Caplan et al1 include in their meta-analysis a trial by Mather et al that compared home care with intensive care management of patients with acute myocardial infarction (AMI) between 1966 and 1968.2 A joint working party of the Royal College of Physicians and British Cardiac Society dismissed the results of this study because of design defects.3,4

Kalra et al5 performed a randomised trial with three arms for patients with acute stroke: stroke unit care, general ward care with stroke team support, and domiciliary care. Stroke units achieved a significantly lower mortality than general ward or domiciliary care. Caplan et al ignore the heterogeneity of the hospital arms, and sum their mortalities, creating a non-existent advantage for domiciliary care over hospital care. This meta-analytic technique is simplistic and invalid.

Hill et al describe home versus hospital management for patients with suspected AMI,2 as do Mather
et al.2 Studies of obsolete treatments, such as home management of patients with AMI, should have been excluded from the meta-analysis.

Rudd et al studied the effect of early discharge after stroke using a 1976 clinical definition of stroke.6 No details of imaging or comorbidities were given. The assumption of equipoise in the trial arms regarding morbidity is not met, and the study is not suitable for inclusion in the meta-analysis.

Indredavik et al7 studied the effect of early supported discharge versus ordinary care in patients with stroke, with 13 deaths at 26 weeks in the experimental group against 15 deaths in the control group. However, Caplan et al incorrectly report this as 21 and 26 deaths, respectively.

If these five most heavily weighted studies are excluded, no significant difference in mortality is seen
(243 hospital-in-the-home deaths [n = 2747] v 245 hospital deaths
[n = 2435], two-sided P = 0.14). Moreover, meta-analysis of the effect of location on mortality where the circumstances of the location are
not defined and not expected to be homogenous is invalid and makes
the mathematical exercise futile.

In reply: Dickson argues for exclusion of randomised controlled trials (RCTs) if treatments have changed, but treatments are constantly changing so, following
this rule, meta-analysis would be impossible. Similarly, diagnosis
has changed — stroke was
a clinical diagnosis, then computed tomography was required, and
now magnetic resonance imaging
is needed. Equipoise is not a requirement for inclusion in a
meta-analysis.

Complaints about research being simplistic because it aggregates patients and groups demonstrates a misconception of research, which is designed to aggregate one factor while other factors differ — for example, study arms may have different mixtures of ages but similar average ages. The meta-analysis studied effects of two systems of care — hospital and hospital in the home (HITH) — not a particular diagnosis or treatment.1,2 Therefore it is legitimate to aggregate hospital patients and compare them with HITH patients.

Location and heterogeneity were mathematically defined, there was no heterogeneity for mortality data, and other outcomes were adjusted appropriately.

Results from the study by Indredavik et al were published in several reports, but (due to space limitations) only the primary report was cited. The data that Dickson refers to are in a report by Fjaertoft et al.3

Although the prevailing opinions
of the Royal College of Physicians
of London and the British Cardiac Society criticised the study by Mather et al in the 1970s, no contradictory facts or trials were cited at the time.4 Considering that other prevailing practices that were initially not examined by adequate RCTs led
to many iatrogenic deaths (eg, prophylactic use of antiarrhythmic drugs5), such practices should be examined and evidence of patient harm taken seriously, rather than simply dismissing evidence as obsolete.