Alarm symptoms in early diagnosis of cancer in primary care: cohort study using General Practice Research Database

For those in Primary Care the diagnostic emphasis is on
distinguishing those patients with serious illness, and requiring further
investigation and management, from the large number with self limiting
conditions requiring little further action. This is especially true in the
case of cancers and the article by Jones et al1 highlights some symptoms
suggestive of cancer.
In this article reference is made to the low sensitivity of rectal
bleeding for the diagnosis of rectal cancer but promulgates the
recommendation for full investigation for all patients with rectal
bleeding.

As over 96% of patients with rectal bleeding do not have cancer2 such an
approach risks exposing a large number of patients to unnecessary and
invasive investigation. Greater awareness of the different symptom
combinations of rectal bleeding3 is a more sensitive approach. The
combination of rectal bleeding with a change in bowel habit to increased
frequency of defecation with or without loose motions, and without
perianal symptoms and increasing age will identify those patients at
higher risk of cancer for more prompt investigation. Patients at lower
risk, with a low level of anxiety regarding their symptoms, can be treated
for longer in Primary Care and investigated if their symptoms fail to
improve or develop into higher risk symptoms.

The identification of high risk symptoms or symptoms combinations could
prove invaluable to those in Primary Care trying to identify the few with
serious underlying disease from the large number of patients seeking
medical advice and safeguard these patients from hospital investigations.

Brian G Ellis, general practitioner, The Swan Surgery, Petersfield,
Hants, GU32 3AB

Iona Heath, general practitioner, Caversham Group Practice, 4 Peckwater
Street, London, NW5 2UP.

Michael R Thompson, Consultant Colorectal Surgeon, Queen Alexandra
Hospital, Cosham, Portsmouth, PO6 3LY.

Competing interests: none

1, Roger Jones, Radoslav Latinovic, Judith Charlton, Martin C
Gulliford, Alarm symptoms in early diagnosis od cancer in primary
care:cohort study using General Practice Research Database, BMJ
2007;334:1040-4

2, Brian G Ellis and Michael R Thompson, Factors identifying higher risk
rectal bleeding in general practice, BJGP 2005; 55: 949-955

3, Thompson M.R. ACPGBI Referral guidelines for colorectal cancer
Colorectal Dis 2002;4: 287-97

Competing interests:
None declared

The paper by Jones et al identifying predictive values for common
presentations of tumour in primary care suggests that the data can
identify patients needing urgent attention. It overlooks an important
question - what level of risk constitutes an alarm symptom?
From a rational perspective, the positive predictive value at which to
investigate should be dictated by economic factors and in particular the
quality adjusted life year (QALY) ultimately gained. For example,
screening for brain tumour in children at a risk of 4% using MRI yields a
cost per QALY of approximately £66,000 (1).
Clearly, many other emotive and complex issues are at play in the decision
to refer particularly where serious pathology may be present.(2) However,
a framework that clarifies the relevance of positive predictive values
remains elusive. Why should 4% be a red flag and not 0.4%? Until one is
forthcoming, an appropriate response to the data that Jones presents is –
interesting but so what?

1 - Medina LS, Kuntz K, Pomeroy S. Children with headache suspected
of having a brain tumor: a cost-effectiveness analysis of diagnostic
strategies.
Pediatrics 2001;108(2):255-63.

2 - Morgan M, Jenkins L, Ridsdale L. Patient pressure for referral
for headache: a qualitative study of GP’s referral behaviour. British
Journal of General Practice 2007;57:29-35.

Competing interests:
None declared

Roger Jones paper on alarm symptoms [1] provides an insight in to the
use of the general practice research database (GPRD).

The size of the GPRD practice database gives precise answers – small
confidence intervals in this case. However, it doesn’t necessarily follow
that the estimates will prove accurate for that population. Readers
should think of this like shooting arrows at a target. Although the arrows
are clustered close together (precision) they may be some distance from
the bulls-eye (accuracy).

The GPRD is a useful tool when the data is proven to be accurately
recorded. For instance drug usage, cancer diagnoses. However, to date no
validation exists for the accuracy of reported symptoms in the GPRD.
Recording and coding in practice varies greatly. General practitioners
tend to record the main problem only [2] which leads to a selective under
reporting and coding of secondary problems.

So what can be inferred from this paper? When GPs perceive the
problem to be an alarm symptom, such as haematuria, but in the presence of
a urinary tract infection, they are more likely to code the latter than
the former due to its prominence as the main diagnoses. Thus the estimates
in the Jones paper are likely to reflect those patients with no
alternative diagnoses; warranting further investigation. The result of the
paper should not be interpreted as a blanket policy for cancer evaluation
of all patients with alarm symptoms such as haematuria when a diagnosis of
UTI or renal calculi is more probable.

[1] Jones R, Latinovic R, Charlton J, Gulliford MC. Alarm symptoms in
early diagnosis of cancer in primary care: cohort study using General
Practice Research Database. BMJ , doi:10.1136/bmj.39171.637106. AE

[2] Britt H, Mezza RA, DelMar C. Methodology of morbidity and
treatment data collection in general practice in Australia: a comparison
of two methods. Fam Pract. 1996 Oct;13(5):462-7.

Competing interests:
None declared

The age specific incidence 0f the relevant cancers is calculable
combining tables 2 and 4.

Table 4 details only the cancer numbers , and the associated PPV. It
is NOT necessary to include the cancer incidence to compute the PPV =
"Proportion of patients with positive test results who are correctly
diagnosed. " .. Table 4 includes the relevant data for the PPV calculation
( [ number of Cancers / number with symptom ], ie: the denominator
population cancels itself).

If you want to compute the actual Cancer incidence, then combine
Table 2's Symptom incidence with Table 4's 'number with Cancer' and
'number with symptom', to derive the denominator population, and divide it
into 'number of cancers'. I think for reasons of space the BMJ may have
culled the data presented, but the essence remains.

Competing interests:
None declared