Association between teenage pregnancy rates and the age and sex of general practitioners: cross sectional survey in Trent 1994-7

Many errors of scientific method seem to be reflected in this
article, not to say some glaring errors about the distinctions between
inference and causality.

“General practices with female doctors, young doctors, or more nurse
time had lower teenage pregnancy rates. The findings may have implications
for the mix of health professionals within primary care.”

This quote implies that a causal link is going to be drawn between
the gender and age of the GP, and the nurse time available, and the number
of teenage pregnancies within that medical practice. On what basis is this
link being made? Many things can be statistically correlated, which have
absolutely no causal association in reality. One thinks of predictions
about sunspots and economic cycles. Though such links can be demonstrated
as strongly correlating, yet no causal link in reality is to be found,
i.e. sunspots do not act as the cause of economic activity, even though
that is an observed pattern of correlation.

Events occurring together often seem to our mind to be causally
linked. A link is suggested, assumed maybe, but not proven. There may be a
third underlying cause for both, or it might just be coincidence. Many
further observations are required before any single observation [=
‘subjective belief’] can be elevated into a more meaningful status. Even
then, it might be coincidental. If we observe two people walking down the
street at the same time every day always together, does that mean the one
walking is the cause of the other walking? No, it does not. It is an
assumption that we have projected onto the observation. We make such
assumptions all the time, but they are not proven. Similarly, in former
times, the ‘spontaneous generation’ of rats, lice and insects in rotting
material, was inferred, from faulty and uncritical observations, and the
absence of knowledge of their spores, eggs, larvae, etc. It was derived
from unbridled inferences founded upon incomplete observations [see
Porter, pp.429-30; Mason, p.293, p.341; Shryock, pp.268-9]. Our minds tend
to run after such inferences, and to draw these imperfect conclusions, but
that is bad science and as a tendency it should be resisted. Well, they
might be good sources of ideas, but can generate strong expectations,
making our perception far from neutral and giving rise to shoddy thinking.

“High conception rates are associated with having a teenage mother,
having divorced parents, poor education, and deprivation. Health
professionals may reduce the harmful effects of deprivation and poor
education on risk of teenage pregnancy by improving access to effective
health education and contraceptive services.”

Again, ‘associated’ in this paragraph suggests ‘is caused by’ and
hence the directive that ‘health professionals may reduce the harmful
effects of deprivation…’ This is an assumption, not a conclusion made from
the data. The assumption being made is that the one is the cause of the
other. On what basis? It is not known, it is inferred.

“We aimed to determine general practice characteristics associated
with variations in teenage pregnancy rates. In particular, we investigated
the effect of the sex and age of the doctor and the availability of a
practice nurse.”

The phrase ‘associated with’ then becomes ‘the effect of the sex and
age of the doctor and the availability of the nurse’. Effect on what? On
teenage pregnancies. Again, we must ask on what basis does this argument
rest? In fact, it is a castle built in the air. It is founded upon the
contention that a causal link has been found from this research. No such
causal link has been found from this research. All that has been found is
a correlation. The rest is conjecture and assumption, projected onto the
data and which might prove to be unfounded.

“Practices with a female doctor, a young doctor, or more practice
nurse time had significantly lower teenage pregnancy rates after
adjustment for other factors. For example, practices with a female doctor
had 91% of the teenage pregnancy rate found in other practices; practices
with a doctor under 36 years had 84% of the rate; practices with both a
female doctor and one under 36 years had 75% of the teenage pregnancy
rate.”

Again, a causal link is being imputed to exist between the patterns
revealed and those factors in the practice, which have been earmarked by
the researchers. Why all the pussyfooting around? They might just as well
come right out with it and say: ‘teenage pregnancies are CAUSED by older
and male doctors, and by less nurse time’. Would that be a fair conclusion
to make? Is this good science or bad science?

“A causal relation, however, cannot be inferred from a statistical
association, particularly in a cross sectional study.”

But we are going to draw them anyway!

“General practices with a female partner have lower teenage pregnancy
rates than those without a female partner. As our study was cross
sectional, we do not know whether female doctors had chosen to work in
areas with low teenage pregnancy rates or whether the presence of a female
doctor influenced such rates.”

This is a supremely illogical paragraph. Do doctors choose to work in
areas with low teenage pregnancies? Is this being realistically presented
as a factor they take into account when CHOOSING a practice? I thought
that doctors more or less ended up where they ended up and had only
limited choice over where they happen to find a post. If this is so, then
the above statement is not only a good example of unbridled and inaccurate
speculation, but also of that wonderful old adage: “lies, damned lies and
statistics”. What is being said may or may not be true, but it has not
been proved by this research

“The association between low teenage pregnancy rate and the presence
of a female doctor may be because female doctors tend to have longer
consultations and handle more problems per consultation and tend to be
more communicative and more patient centred. Female doctors report less
difficulty in discussing sexual problems with teenagers and are more
likely to provide information about the prevention of sexually transmitted
diseases and the use of condoms. Our findings might be due to differences
in case mix between practices with and without a female doctor since
patients presenting to female doctors tend to be younger, more often
female, have female specific problems than those presenting to practices
without a female doctor.”

They say they are not going to causally link these things and then do
precisely that. Why say ‘our findings might be due’? Due means cause. This
paragraph reads like a blatant attempt to explain the events shown in the
data. Which is what they claim [earlier on] not to be trying to link
causally. Which way do they want to play it?

“Practices with a young doctor have lower teenage pregnancy rates
than those without a young doctor. Although younger doctors may be more
interested in teenage health issues, little is known about the effect of
the age of the doctor on provision of teenage contraceptive services.”

‘The effect’ is another causal term. This is another blatant attempt
to ascribe causality to the data. Why say they are not going to ascribe
cause and then go and do exactly that?

“Practices with more practice nurse time had significantly lower
teenage pregnancy rates than those with less practice nurse time. Since
10% of all practice nurse consultations are with teenagers and up to 3% of
all nurse consultations are for contraceptive advice, there may be scope
for further developing the practice nurse's role in the delivery of
contraceptive services to teenagers. We are unable to explain the
association between fundholding and higher teenage pregnancy rates.”

They show how the ‘role of the nurse’ might impinge upon teenage
pregnancies. Again, a causal relationship [‘explain the association’] is
being summoned. If we call ‘low teenage pregnancies’ ‘B’, then we can
identify in this article several factors, which are alleged to be causes
of B. These are as follows:

A = nurse time
C = female doctors
D = young doctors

Thus, a deduction being made is that B is a product of A and C and D.

Likewise, they assert that the following are causes of high teenage
pregnancies [H]:

E = less nurse time
F = male doctors
G = older doctors

They make these assertions because they have been statistically
correlated. Yet, common-sense might suggest that these 'causal links' seem
nebulous at best. Thus, the attempts to explain the patterns in the data
appear feeble, unconvincing, fundamentally flawed and probably futile.
Much deeper and more critical thinking should have been engaged in before
it was decided that these ‘deductions’ were proven by this research and
worthy of publication.

Surely, the authors of reports of this kind owe the public a duty of
much greater care in the somewhat dubious deductions they are placing into
the public domain about such an emotive subject? Ideas can be very easily
misinterpreted. It would seem that some help with scientific method
relating to causation and inference might not have gone amiss. This type
of research might even be seen as bringing medicine into disrepute, by
appearing shoddy, unrigorous and unscientific. Much glitters that is not
gold; genuine scientific truths are not so easily won. Like the
‘spontaneously generated’ organisms of old, the assertions made by this
article seem to stand as a testimony to over-inference from insufficient
observations, very uncritically evaluated. And would on those grounds be
castigated as ‘bad science’.

This might seem somewhat off-topic, and I was preparing this quote
for something else, but it seems highly relevant:

'Kant was saying that…the immediate objects of perception depend not
only upon our sensations but also on our perceptual equipment which orders
and structures those sensations into intelligible unities…some of the
properties we observe in objects are due to the nature of the observer
rather than the objects themselves. There are basic concepts [or
categories] like cause and effect, which are not learnt from experience
but constitute our basic conceptual apparatus for making sense of
experience and the world.' [Chamber's, p.807]

I think that is fair comment on this article.

Peter Morrell

References

Chamber's Biographical Dictionary, 1996

Mason, S F, 1953, A History of the Sciences, RKP, London

Porter, Roy, 1998, The Greatest Benefit to Mankind, A Medical History of
Humanity, Norton, New York & London

Shryock, Richard H, 1936, The Development of Modern Medicine, Univ.
Pennsylvania Press, USA

In the "key messages" box of their article, Hippisley-Cox et al (1)
concluded that general practices with a female and young doctors had
significantly lower teenage pregnancy rates and advised those responsible
for recruiting staff in primary care to take these conclusions into
account. Their arguments are flawed.

There are several methodological problems. First, the authors did
not take into account the doctors' inclusion in the "family planning
list", their possession of family planning certificates and the DRCOG
qualifications. Female and young doctors may be more likely to have
undergone family planning training and obtained these qualifications.
These are key confounding factors as the doctors' skills in providing
contraception were under study. Whether the practice ran a "shared list"
system in allocating patients to doctors is also important. Second,
whereas teenage conceptions are defined as those from girls aged 13 - 15
in the Health of the Nation targets (2) and in other studies, the authors
included all pregnancies from girls aged 19 or under. Third, as a large
proportion of "teenage pregnancies" in this study were from girls aged 17
- 19 who were likely to be in further or higher education during term
time, terminations of pregnancies may have been performed outside Trent
region. Fourth, the incidence rate ratio for the presence of female
doctors was adjusted from 0.84 to 0.94 after taking into account Townsend
score and practice characteristics. However, as good general education is
the most important factor associated with deferring pregnancy (3),
Townsend score alone is unlikely to have fully corrected for this factor.
Finally, the authors properly acknowledged that causation could not be
concluded from a cross-sectional survey

Even if the conclusions were valid, the clinical significance for
individual practices is minimal. For a 5-partner practice with 10,000
patients, there may be approximately 250 girls aged 13 - 15. Taking the
highest teenage conception rate in UK as 16 per 1000 (2), one may expect
to see 4 teenage pregnancies in a year. Even if it were true that
practices with both a female and a young doctor had 75% of the teenage
pregnancy rate, one would only expect one teenage pregnancy to be avoided
a year. These results fall far short of the stringent statutory
requirements (4) to argue that sex is a genuine occupational qualification
to justify sex discrimination.

References

1 Hippisley-Cox J, Allen J, Pringle M, Ebdon D, McPhearson M,
Churchill D, Bradley S. Association between teenage pregnancy rates and
the age and sex of general practitioners: cross sectional survey in Trent
1994-7. BMJ 2000; 320: 842-845.

2 Adler M. Sexual health - health of the nation failure. BMJ 1997;
314: 1743-1748.

3 NHS Centre for Reviews and Dissemination. Effective Health Care.
Preventing and reducing the adverse effects of unintended teenage
pregnancies, vol. 3. York: University of York, 1997:1-11.

4 Sexual Discrimination Act 1975 section 6(1)