Around a quarter of individuals in developed nations succumb to cancer, a figure that is unfortunately on the rise. Despite advancements in screening, the majority of cancers are initially detected when patients present to primary care physicians with symptoms. General Practitioners (GPs), therefore, are at the forefront of cancer diagnosis, tasked with the critical challenge of differentiating between symptoms indicative of malignancy and benign conditions. This review delves into the intricate process of early cancer diagnosis within primary care, focusing particularly on the strategies for selecting patients who require urgent investigation. We will concentrate on the four most prevalent cancers in the UK – breast, lung, colon, and prostate – as these have been the primary focus of recent research and hold significant implications for early detection and improved patient outcomes.
INTRODUCTION
Cancer remains a leading cause of mortality in the developed world, accounting for approximately 25% of deaths. This proportion is increasing as other causes of death are being effectively addressed. Crucially, most cancer diagnoses begin with patients presenting symptoms to their primary care physician. While the diagnosis of a new cancer is not a daily occurrence for GPs (in the UK, a full-time GP might expect to diagnose a new cancer in a patient roughly once a month), the consideration of cancer as a potential diagnosis is a routine aspect of their practice. GPs, through repeated exposure, develop substantial expertise in ruling out cancer and, over time, in identifying it. However, a significant point of frustration for GPs is the disproportionate emphasis placed on their diagnostic performance in cases where cancer is ultimately confirmed, with limited recognition for their adeptness in correctly diagnosing the far more numerous cases where cancer is not present.
Over the past decade, advancements in diagnostic pathways, particularly the introduction of rapid investigation clinics in the UK offering specialist assessment within two weeks, should theoretically have streamlined cancer diagnosis. National guidelines exist to direct patient selection for these clinics, although adherence to these guidelines can be inconsistent. Furthermore, access to primary care imaging has improved in terms of both availability and speed. Despite these advancements, the UK’s cancer mortality rates remain comparatively high. Some argue that these new facilities have not yielded significant improvements in overall outcomes. Current estimates suggest that between 7,500 and 10,000 lives are lost annually in the UK due to delays in cancer diagnosis. This figure, however, encompasses various stages of the diagnostic process, as illustrated in Figure 1.
Figure 1.
Intervals used in the literature of diagnostic delays.
Much of the research to date has examined the entire diagnostic interval from point A to E (Figure 1), with fewer studies specifically dissecting where delays occur and where the greatest opportunities for improvement lie. This review will primarily focus on the interval between B and D, from the initial presentation of symptoms to referral for definitive investigation. The examples discussed will largely be drawn from the four most common cancers: breast, prostate, lung, and colorectal cancer.
Patients can typically recall when their symptoms began and when they first consulted their GP. However, at the initial consultation, it may not be immediately apparent to the GP that these symptoms are indicative of cancer. Evidence suggests that patients themselves play a crucial role in filtering symptoms. Many symptoms that could potentially signify cancer are not reported to primary care because patients often correctly self-assess that there is a benign explanation that does not necessitate medical intervention. For example, only 41% of individuals experiencing rectal bleeding seek medical advice. Therefore, when a patient presents with a symptom, there are two key selection stages: first, the patient’s decision to seek medical attention, and second, the GP’s decision to refer for further investigation. Importantly, the probability of cancer is higher in those who have progressed through both selection stages, demonstrating that the selection process, while imperfect, is functional. Interestingly, the relative increase in cancer risk between the first selection stage (deciding to consult) and the second stage (GP referral decision) is notable, with the first stage demonstrating a greater increase. To illustrate, the positive predictive value (or risk) of colorectal cancer following a single episode of rectal bleeding in the general community is approximately 0.1%. Once this bleeding is reported to primary care, the risk escalates to around 2-3%. In patients subsequently referred to secondary care, the risk further increases to 5-7%.
How this fits in
Recent years have witnessed a surge in research focused on cancer diagnosis, particularly within the primary care setting. The selection of patients for investigation has always been a complex aspect of general practice. This review aims to summarise recent advancements in this field, with a specific focus on evidence supporting a risk-based approach to referral strategies in primary care.
This escalating gradient of risk highlights the critical ‘gatekeeper’ function of primary care, where GPs are responsible for selecting patients for further investigation or specialist referral. One perspective suggests that the UK’s suboptimal cancer outcomes might stem from GPs having an excessively high threshold for referring patients for investigation. Paradoxically, while specialist clinics often criticize GPs for not adhering to referral guidelines, they simultaneously acknowledge the limited accuracy of these guidelines in identifying cancer. This creates a challenging situation for GPs, who must navigate patient expectations while also ensuring the judicious use of healthcare resources. The following sections of this review will summarise recent progress in primary cancer diagnosis, aiming to provide insights that can aid GPs in this delicate balancing act.
Breast Cancer
Surprisingly, for such a prevalent and emotionally charged cancer, there is a relative scarcity of primary care-based evidence to guide GPs in their referrals of women presenting with breast symptoms. While extensive research exists on breast cancer screening and genetics, a disproportionate amount of primary care research focuses on GP adherence to referral guidelines rather than on the evidence underpinning the formulation of these guidelines. It is important to note that the majority of breast cancers are not detected through screening programs; in fact, approximately three-quarters are diagnosed following symptomatic presentation to primary care.
Breast symptoms account for nearly 3% of all female consultations in primary care settings. The most frequently reported symptoms are breast pain and breast masses, with these concerns being most common in women aged 25-44 years. Nipple discharge or changes in breast skin texture are reported much less frequently. A breast lump carries a likelihood ratio of 15 for cancer, while nipple complaints have a likelihood ratio of 3.1, and breast pain, 1.8. Box 1 outlines the two key metrics used in diagnostic research. Overall, approximately 8% of women presenting to primary care with a breast lump are subsequently diagnosed with breast cancer. As expected, this figure is significantly influenced by age, with breast cancer incidence being 15 times higher in women aged 45-64 compared to those under 25. Consequently, in women under 25, the risk of cancer associated with a breast lump is approximately 0.5%, assuming likelihood ratios remain relatively consistent across age groups, as is generally observed in cancer diagnostic studies. Less than 2% of women with nipple complaints are diagnosed with cancer, and the percentage is even lower for breast pain, at 0.9%. These statistics, published after the introduction of NICE guidelines, support the concept of incorporating an age threshold for urgent referral of breast lumps. Indeed, the prominence given to nipple bleeding in referral guidelines is questionable, as only a small proportion of women with this symptom have underlying cancer. Similarly, breast pain is genuinely a very low-risk symptom for malignancy.
Box 1. Metrics used in diagnostic research
Likelihood ratio. This ratio represents the odds of a patient with cancer exhibiting a specific symptom divided by the odds of a patient without cancer having the same symptom. For example, in lung cancer, 20% of patients report haemoptysis (coughing up blood) to their doctor, whereas 1.5% of ‘healthy’ individuals do so. In this case, the positive likelihood ratio is 20/1.5 = 13.
Positive predictive value. This value indicates the probability of a patient having the disease of interest when they present with a particular symptom. Using the same lung cancer example, a person reporting haemoptysis to their GP has a 2.4% chance of the symptom being caused by underlying lung cancer.
These two metrics are related. The positive predictive value of a symptom is derived from the incidence of the relevant cancer multiplied by the likelihood ratio for that symptom. The actual calculation is slightly more complex, requiring the conversion of positive predictive value and incidence to odds, and then back to percentages. Furthermore, the timeframe for symptom relevance must be considered (e.g., a cough occurring more than a year before a lung cancer diagnosis is unlikely to be related). Nevertheless, the core principle is that a higher likelihood ratio corresponds to a higher positive predictive value. Conversely, a lower cancer incidence (rarer cancers) results in a lower positive predictive value.
In younger women, GPs must rely heavily on their clinical judgment to identify suspicious breast lumps. However, many women in this age group may be referred simply to provide adequate reassurance. Younger women are at a higher risk of inflammatory breast cancer (which accounts for approximately 5% of all breast cancers). As the name suggests, the primary features are breast swelling, redness, and warmth. ‘Peau d’orange’ skin changes (skin resembling orange peel) are also common. These symptoms often develop rapidly, making differentiation from infection challenging. The underlying pathology involves blockage of lymphatic vessels by cancer cells, resulting in a poorer prognosis for this type of breast cancer.
Lung Cancer
Lung cancer diagnosis is often associated with a sense of nihilism and fatalism. The overall prognosis remains poor, leading some to question the value of early diagnosis. Patients themselves may share this pessimism, sometimes due to feelings of guilt related to smoking history, which can contribute to delayed symptom presentation. By the time symptoms are reported, the cancer has often metastasized, rendering curative treatment unlikely. Furthermore, lung cancer symptoms are typically non-specific, making early identification challenging even when patients present promptly. These factors contribute to the fact that only approximately 20% of UK lung cancer patients are eligible for surgical resection, with only 17% actually undergoing surgery, and around half of these surviving for 5 years.
Currently, lung cancer screening offers limited benefit. Conventional sputum cytology is ineffective due to low sensitivity, and plain chest X-ray screening is also not beneficial. The UK Lung-SEARCH trial is currently investigating annual cytology, enhanced by computer-assisted image analysis, in smokers with chronic obstructive pulmonary disease. Two large screening trials using spiral CT are ongoing, with the US trial expected to report final results in 2010 and the European trial later in that decade. However, 3-year results from a smaller Italian trial were discouraging. Spiral CT can detect nodules as small as 0.5 cm, compared to the 3 cm detection limit of plain radiography. However, the ability to detect such small nodules is accompanied by a high false-positive rate: 21% after one scan and 33% after two. The false-positive rate may be lower in the Lung-SEARCH trial due to the highly targeted population and higher incidence of lung cancer in this high-risk group.
Given the often lengthy period between symptom onset and presentation to a doctor, efforts have been made to improve symptom awareness and encourage earlier presentation. Such initiatives are difficult to evaluate, but audits suggest a modest effect in terms of earlier stage diagnosis and increased resectability. A campaign in Doncaster, UK, encouraged patients with cough to report it to their GP, coupled with more lenient criteria for chest X-ray requests. This campaign resulted in an increase in the percentage of cancers diagnosed at stages I or II, from 11% to 19%, a statistically significant improvement (P = 0.02). It is likely that progress in lung cancer outcomes will come from such interventions, rather than from screening programs, although many lung cancers will unfortunately remain incurable.
Lung cancer diagnosis has one advantage over many other cancers: the primary diagnostic test, a chest X-ray, is readily accessible, relatively inexpensive, and reasonably accurate. Approximately one in five chest X-rays requested in primary care are for suspected lung cancer. However, chest X-rays can be negative in about a quarter of lung cancer cases, due to lesions being missed by radiologists or being non-visible on X-ray. Even so, a 25% false-negative rate is relatively low for a primary care cancer test. False-positive chest X-rays are also relatively infrequent (reflecting the 3 cm minimum nodule size for reliable identification), although many X-rays reveal non-specific abnormalities requiring repeat testing. Therefore, a low threshold for requesting a chest X-ray in patients with symptoms suggestive of lung cancer is justifiable. Broadening the criteria for chest X-ray requests would likely yield a small increase in cancer detection. Unpublished estimates from a UK primary care cohort examined a hypothetical policy of X-raying all current smokers over 60 presenting to primary care with a cough. Such a policy could expedite diagnosis in approximately 25% of lung cancer cases by a median of 9 months, potentially improving mortality. However, the ‘cost’ would be borne by individuals without cancer, with an estimated number needed to X-ray of 47 to detect one of these 9-month expedited cancers. This may represent an efficient use of resources compared to the costs and false-positive rates associated with screening programs.
For patients presenting to primary care with symptoms, recent evidence provides a basis for investigation and referral decisions. Haemoptysis (coughing up blood) is a significant symptom, carrying a 2.4-7.5% risk of lung cancer depending on age and sex. This clearly warrants a chest X-ray. However, haemoptysis is relatively uncommon in lung cancer, reported by only 40% of patients presenting to primary care. Other symptoms, including cough, dyspnoea (shortness of breath), loss of appetite, weight loss, fatigue, and chest pain, are more prevalent in lung cancer than haemoptysis. However, they are even more common in benign conditions, resulting in positive predictive values of less than 1% for lung cancer. Positive predictive values are higher in smokers or those over 70, ranging from 1-2%. Arguably, these higher-risk groups warrant a chest X-ray even at this lower risk threshold, given the accessibility and low cost of chest X-rays. For other cancers where diagnostic tests are more complex, the threshold risk value for investigation remains a subject of debate, with varying perspectives among patients, clinicians, and healthcare commissioners. Despite the non-specific nature of symptoms in most lung cancer cases, the median GP delay (interval B-D in Figure 1) in the UK is 51 days, compared to 26 days in Sweden. As with most cancers, the majority of symptomatic patients are identified and investigated relatively promptly, while a minority, albeit a significant one, experience considerable delays.
Colorectal Cancer
Colorectal cancer has been the subject of considerable research attention recently, including published systematic reviews incorporating both primary and secondary care data, and primary care-focused reviews. Several recent studies have examined rectal bleeding. No study has reported a rectal bleeding risk below 2%, except in individuals under 60 years of age. Current UK referral guidelines recommend urgent referral for rectal bleeding persisting for 6 weeks or accompanied by diarrhea. However, the 6-week waiting period before a 2-week urgent referral is illogical. The risk associated with rectal bleeding alone in individuals over 60 justifies investigation, regardless of accompanying symptoms.
While guidance on rectal bleeding is questionable, the guidelines regarding iron-deficiency anemia are demonstrably flawed. The haemoglobin level threshold for rapid colorectal cancer investigation in cases of iron-deficiency anemia is 10 g/dL in women and 11 g/dL in men. These thresholds correspond to colorectal cancer risks of 13.3% and 7.7% respectively. Clearly, patients with milder levels of anemia also warrant investigation for potential colorectal cancer, especially given that anemia is a colorectal cancer symptom associated with poorer mortality.
Referral decisions become significantly more challenging when considering low-risk-but-not-no-risk symptoms such as abdominal pain, constipation, or diarrhea. These symptoms have positive predictive values of 1% or less for colorectal cancer. In lung cancer, advocating for increased investigation of low-risk respiratory symptoms is feasible due to the accessibility and characteristics of chest X-rays. However, colonoscopy, while highly effective, is more costly and invasive. CT colonography was initially hoped to replace colonoscopy, but it misses approximately 1 in 10 lesions larger than 1 cm, which are reliably detected by colonoscopy. Future possibilities include biomarkers, such as matrix metalloproteinase-9, serum levels of which correlate with colorectal cancer presence, and free DNA detected in the rectum of some colorectal cancer patients. However, further research is needed before these can be implemented in primary care. Simpler approaches include scoring systems for low-risk symptom patients, such as the CAPER and SELVA scores. Both systems face similar challenges: achieving high sensitivity (identifying most cancer cases) necessitates sacrificing specificity, leading to many patients undergoing negative investigations despite being classified as ‘higher risk’ by the scores.
Until more effective strategies emerge, GPs must rely on subtle clinical cues to identify colorectal cancer in patients with low-risk symptoms. This is achievable and often done effectively. For example, the term ‘change in bowel habit’ strictly refers to constipation or diarrhea. However, in clinical practice, it carries a broader connotation. In the UK, when a GP uses this term, it implies ‘constipation or diarrhea, with a reasonable suspicion of colorectal cancer.’ Few patients with documented ‘change in bowel habit’ are not investigated for cancer, while many patients with documented diarrhea, for instance, are not investigated. And GPs are often correct in this distinction: the positive predictive value of ‘change in bowel habit’ in an 80-year-old man is 4%, compared to 1.2% and 0.7% for diarrhea or constipation alone in the same age group. Some additional, yet undefined, features enable GPs to identify these higher-risk patients, who are at least three times more likely to have cancer. The nature of these features and how GPs detect them remains unclear. It is unlikely that GPs consciously employ Bayesian probability calculations, but subconsciously, this may be the underlying mechanism.
Prostate Cancer
Similar to lung cancer, there is a degree of nihilism surrounding the value of early prostate cancer diagnosis, albeit from a different perspective. The argument, crudely stated, is that early lung cancer diagnosis is futile because the patient will likely die regardless, while early prostate cancer diagnosis is unnecessary because the patient will likely survive anyway. Much of this nihilism stems from prostate cancer screening trials, which have shown limited evidence of mortality benefit. Treatment of small, localized prostate cancers yields a small mortality gain but carries the risk of complications such as incontinence. However, treatment of larger tumors and metastatic disease is less controversial. Larger tumors are also more likely to cause symptoms, simplifying the diagnostic challenge in primary care. Given the benefit of treating symptomatic prostate cancer, it is appropriate to investigate for prostate cancer in men presenting with lower urinary tract symptoms (LUTS). However, LUTS indicate prostate enlargement but do not differentiate between benign prostatic hyperplasia (BPH) and malignancy. Most common LUTS have positive predictive values for prostate cancer of around 3%. Impotence can also be an early symptom of prostate cancer, with a similar risk of approximately 3%. Therefore, the clinical management of men with LUTS should include digital rectal examination (DRE) and prostate-specific antigen (PSA) measurement. This use of PSA differs significantly from its use as a screening tool. While a small percentage of men with prostate cancer may have a PSA level as low as 1 ng/mL, these cancers typically have a favorable prognosis, and missing them is of limited clinical consequence. Urologists sometimes use PSA velocity (rate of PSA increase) or free PSA levels to guide biopsy decisions. However, evidence to support these measurements in primary care is lacking.
Ovarian Cancer
Recent primary care-based studies have shed light on ovarian cancer, a notoriously challenging malignancy. It is now clear that the term ‘silent killer’ is inaccurate. Primary care studies consistently demonstrate that symptoms are common and generally reported to GPs. Ovarian cancer is, in fact, a ‘noisy killer.’ The difficulty lies in the non-specificity of the symptoms, which often include fatigue, abdominal pain, and urinary frequency in the early stages. Abdominal distension carries a relatively higher risk of cancer, with a positive predictive value of 2.5%, despite being omitted from current referral guidelines. One issue appears to be that GPs, understandably, may not consider ovarian cancer in the differential diagnosis for these symptoms, or may deem it too improbable to warrant examination or investigation. Consequently, ovarian cancer is often perceived as a cancer frequently missed by GPs. However, GPs are likely no worse (or better) at diagnosing ovarian cancer than other internal malignancies. Serum CA125 measurement is available in primary care and has reasonable specificity, but its sensitivity is unknown, making its value debatable. Transvaginal ultrasound is a more promising test, extensively evaluated in screening studies with good performance characteristics. However, extrapolating from screened populations to symptomatic populations carries inherent risks. A large UK screening trial has completed recruitment, but mortality follow-up and main results are expected after 2014.
Other Cancers
Research in primary care for other, less common cancers is limited, largely due to their rarity. Referral decisions for these cancers often rely on GP experience and intuition, with intuition likely playing a more dominant role. Limited studies generally support current standard clinical practice. For instance, the risk of esophageal cancer in patients presenting with dysphagia (difficulty swallowing) is 5.7% in men and 2.4% in women, and the risk of urinary tract malignancy in patients with hematuria (blood in urine) is 7.4% and 3.4% respectively, suggesting that investigation is warranted in these cases. Similarly, GPs appear proficient at identifying patients with malignant cervical lymphadenopathy (swollen lymph nodes in the neck) and, importantly, at distinguishing those with benign causes. Overall, less than 2% of cervical lymphadenopathy cases presenting to primary care are malignant. The risk of brain tumor in patients with headache is approximately 1 in 1000, supporting the view that routine brain scanning for typical headaches is not necessary.
Summary
Primary care is the central setting for cancer diagnosis or suspicion. It is also the locus for pre-symptomatic risk assessment, including family history compilation and addressing modifiable risk factors like smoking and obesity. Cancer as a diagnostic possibility is a near-daily consideration in clinical practice. The evidence base guiding patient selection for referral, and equally importantly, for reassurance and non-investigation, is continuously expanding. At times, this evidence may contradict current guidelines, and in such instances, most GPs rely on their clinical acumen, often correctly. The majority of cancer patients receive effective primary care services. Future research should focus on identifying strategies to detect the minority of patients who experience diagnostic delays at an earlier stage, thereby improving outcomes and survival rates in this critical area of healthcare.
Acknowledgments
Peter Wood provided valuable critique and suggestions on an earlier draft of this manuscript.
Linked commentary
This article has an associated commentary: Neal RD. Cancer diagnosis — the role of urgent referral guidelines. Br J Gen Pract 2010; DOI: 10.3399/bjgp10X483427.
Funding
No specific funding was received for this work. William Hamilton is supported by a NIHR postdoctoral fellowship. The views expressed are those of the author and do not necessarily reflect those of the Department of Health.
Competing interests
William Hamilton has received research funding from various sources, including NIHR School of Primary Care Research, the Department of Health, CRUK, and Colonix Ltd. None of these funding sources influenced the content of this review.
Discuss this article
Contribute and read comments about this article on the Discussion Forum: http://www.rcgp.org.uk/bjgp-discuss
