Cancer risk in persons with new-onset anaemia: a population-based cohort study in Denmark

We performed an observational population-based cohort study using electronic data from Danish laboratory information systems [21] and nationwide healthcare registries [22‐24]. The unique civil personal registration (CPR) number (assigned to all Danish residents) allows accurate individual-level linkage of these data (Fig. 1). The laboratory systems hold information on all blood test results requested in general practice or at hospitals and analysed at a department of clinical biochemistry. The Cancer Registry holds information on all cancers diagnosed in Denmark, including cancer type and time of diagnosis. The National Patient Register and the Psychiatric Central Research Register hold information on all hospital contacts and diagnoses in Denmark. The Civil Registration System and Statistics Denmark hold demographic and socioeconomic data on all residents in Denmark.

This study is based on data from persons living in the North Denmark Region (0.6 million inhabitants) or the Central Denmark Region (1.3 million inhabitants), which are two of the five Danish regions. We used the laboratory systems in these two regions, as these comprise complete laboratory data within the study period (in contrast to the nationwide laboratory database) [21, 25]. Danish residents (5.8 million inhabitants) listed with a specific general practice (99%) have unrestricted access to healthcare free of charge, and the general practitioner (GP) serves as gatekeepers and coordinators to the specialised healthcare (except for emergencies, ear-nose-throat specialists, and eye specialists) [22, 26].

The inclusion criteria were: (i) 40–90 years of age at the date of anaemia (as we considered this age group to be clinically relevant for considerations of cancer), (ii) new-onset anaemia detected by a blood test requested from general practice and recorded in the laboratory systems, (iii) date of new-onset anaemia between 1 April 2014 and 31 December 2017, and (iv) living in one of the two included regions at inclusion.

We did not allow persons to re-enter the cohort, and we excluded persons with a prior history of cancer recorded in the Cancer Registry.

We estimated the time to a cancer diagnosis from the date of new-onset anaemia. We assessed the incidence proportion of cancer based on the Aalen-Johansen estimator, considering death as a competing risk. We used the Aalen-Johansen estimator instead of the Kaplan Meier estimator as we considered death as a competing event, and Kaplan-Meyer estimates are biased on data with competing risks [38]. We followed all persons until a cancer diagnosis, death, emigration from the included regions, or end of 12- month follow-up, whichever came first. We stratified analyses by anaemia type and sex.

Further, to compare the cancer incidence in the study population to a general population, we estimated standardised incidence ratios (SIRs) of cancer and of specific cancer types based on age- and sex-specific cancer incidence rates in a general population by use of the NORDCAN database [39, 40]. NORDCAN is a database of cancer statistics for the Nordic countries, and includes information on e.g. cancer incidence [39, 40]. We included a general Danish population aged 40 to 85+ years, and used estimated 12-month cancer incidence rates in a comparable period from 2014 to 2018 [39, 40]. We stratified the analyses by anaemia type and sex.

To identify persons at increased risk of cancer, we estimated the associations between patient characteristics and risk of cancer. We assessed the hazard ratios (HRs) by applying a multivariable Cox proportional hazard model, treating competing risk (death) as censoring. We evaluated the proportional hazard assumption from log-minus log plots, and we detected no violation of the assumptions. We followed the persons until any first cancer diagnosis, emigration from the included regions, death, or end of 12- month follow-up, whichever came first. We stratified the analyses by anaemia types and adjusted for age (continuous), anaemia severity, civil status, educational level, income, comorbidity, and sex. Missing values occurred in the variables with educational level (n = 1979, 4.0%) and income (n = 203, 0.4%), and were handled as representing ‘low educational level’ and ‘low income’. All persons (n = 48,925) were registered in the Civil Registration System, and thereby no loss to follow-up.

A total of 7.9% (CI = 7.6–8.2) of men and 5.2% (CI = 4.9–5.5) of women were diagnosed with cancer within 12 months. Gastrointestinal cancer was the most frequent cancer type in both men (2.7%, CI = 2.5–2.9) and women (2.2%, CI = 2.1–2.4).

Across the anaemia types, the highest cancer incidence proportion was seen in men with CIIDA (19.3%, CI = 14.5–24.1) and women with AI (15.3%, CI = 13.1–17.5) (Fig. 3a, Fig. 3b). Gastrointestinal cancer was the most frequent cancer type in persons with IDA, CIIDA, AI (women) and unclassified anaemia; the highest proportion was seen in men with CIIDA (10.8%, CI = 7.0–14.6). Respiratory system cancer was the most frequent cancer type in men with AI (5.0%, CI = 3.9–6.0), whereas haematological cancer was the most frequent cancer type in persons with ‘other’ anaemia (men 1.8%, CI = 1.3–2.3, women 1.9%, CI = 1.2–2.6).

Most cancers were diagnosed 3–6 months after the new-onset anaemia (Fig. 3a, Fig. 3b).

The SIRs for overall cancer was 5.1 (CI = 4.9–5.3) in men and 4.1 (CI = 3.9–4.4) in women (Fig. 4). The cancer type with the highest SIR was haematological cancer in both men (SIR 11.5, CI = 10.4–12.8) and women (SIR 11.1, CI = 9.6–12.9).

Across the anaemia types, the highest SIR for overall cancer was seen in men with CIIDA (SIR 12.3, CI = 9.4–16.3) and in women with AI (SIR 12.1, CI = 10.3–14.1). Across the cancer types, the highest SIR was seen for respiratory system cancer in men with AI (SIR 26.6, CI = 21.4–31.1) and haematological cancer in women with AI (SIR 38.1, CI = 26.3–55.2) (Fig. 4).

Persons aged 70–79 years were more likely to get cancer compared to persons aged 40–49 years (HR 9.32, CI = 7.64–11.37) (Fig. 5); the highest likelihood was seen in persons aged 70–79 years with IDA (HR 18.32, CI = 11.96–28.09) (Table 2). Persons with severe anaemia were more likely to get cancer compared to persons with mild anaemia (HR 5.17, CI = 4.41–6.06); the highest likelihood was seen in persons with unclassified severe anaemia (HR 6.47, CI = 5.10–8.21). Women were less likely to get cancer compared to men (HR 0.58, CI = 0.54–0.63); the lowest likelihood was seen in women with IDA (HR 0.44, CI = 0.36–0.54). Persons with three or more comorbidities were less likely to get cancer compared to persons without comorbidity (HR 0.58, CI = 0.54–0.63) (Fig. 5); the lowest likelihood was seen in persons with CIIDA and three or more comorbidities (HR 0.25, CI = 0.10–0.64) (Table 2).

As a supplement to the HRs, number of cancer cases stratified by patient characteristics and anaemia types are shown in an Additional file (see Additional Table 1).

This population-based cohort study of nearly 49,000 persons with new-onset anaemia detected in general practice revealed a risk of cancer in 7.9% of men and 5.2% of women within 12 months. Around one in six persons with ‘anaemia of inflammation’ or ‘combined inflammatory iron deficiency anaemia’ got a cancer diagnosis within 12 months. About one in ten of the cancers occurred in the group with IDA. Thus, the majority of cancers came from outside this group traditionally investigated for gastrointestinal cancer. The cancer incidence increased particularly in the first 2–3 months after the anaemia date followed by a significantly slower increase for some cancer types, which could indicate diagnostic activity. For some of the anaemia types, the cancer incidence continued to increase during all 12 months after the anaemia date.

Among the new-onset anaemias, a four- to five-fold higher cancer incidence occurred in women and men compared to the general population. Among the anaemia types, an 11- to 12-fold higher overall cancer incidence occurred in men and women with ‘anaemia of inflammation’ or ‘combined inflammatory iron deficiency anaemia’ compared to the general population. Across cancer types, these two anaemia types had a 10- to 30-fold higher cancer incidence compared to the general population.

This large-scale population-based cohort study holds individually linked data from nationwide registries and laboratory systems known for a high validity and completeness [21, 24]. This ensured virtually complete follow-up with limited risk of selection bias and information bias [21, 24]. Further, the general practice setting, including an unselected population with free access to healthcare services, [22] makes the results widely relevant and may be generalized to other countries with similar access to the healthcare system and with comparable populations (it may not be generalized to other socioeconomic or geographical settings with a higher prevalence of anaemia related diseases, e.g. Thalassaemia in Middle East countries or malnutrition in Africa).

We lacked information on potential and unavailable confounding factors, e.g. smoking and obesity. Thus, we cannot rule out the potential of residual confounding. Further, in the multivariable Cox regression analysis, there is a potential risk of overfitting of the analysis in the smallest group with CIIDA. Furthermore, we had no information on the reasons for the persons to consult their GP, and we do not know what prompted the GP to investigate for an anaemia. Thus, this may introduce confounding by indication as persons having encounters with their GP and having blood tests performed are likely to be more ill compared to the general population. However, haemoglobin measurement is one of the most frequently performed blood tests [41]. This may indicate that non-specific or opportunistic screening for anaemia may be the indication in many cases. If so, this may indicate that this confounding factor may be less dominant. The Cox model treated competing risk (death of any cause) as censoring. However, this was likely informative censoring which could bias the results. Moreover, detection bias might have occurred because clinicians may be aware of anaemia as a sign of cancer, and this paradox might have led to diagnostic evaluation and cancer detection. However, previous research has revealed that the laboratory and diagnostic process of new-onset anaemia in general practice is suboptimal [36, 37]. Therefore, this may reduce the potential risk of detection bias. All these factors imply that we regard the findings as exploratory rather than causal.

The unselected population and the significantly increased cancer risk after new-onset anaemia make it important to establish the aetiological reason for the anaemia. Still, the majority of new-onset anaemia cases were unclassified, [36] and it is unknown which anaemia type these may represent. The proportion of unclassified anaemias in other countries is unknown and has not previously been included when exploring the cancer risk in selected anaemia types. Additional research is needed to explore this large group of persons having unclassified anaemia in general practice.

To our knowledge, this is the first large-scale study to investigate the overall risk of cancer and the risk of specific cancer types across selected anaemia types detected in a general practice population with new-onset anaemia, including CIIDA and unclassified anaemia.

Previous studies have mainly focused on anaemia/IDA and gastrointestinal cancer [10‐13]. These have reported a risk of gastrointestinal cancer of 1–10%, which is in accordance with our findings [10‐13]. Additionally, the one-year SIR for overall cancer in persons with IDA has previously been reported to be 6.12 (CI = 5.57–6.78) in men and 5.60 (CI = 5.13–6.11) in women [18]. Compared to our findings, these figures are slightly lower in men and slightly higher in women. This could be due to different standard populations and other definitions of IDA. Another study reported a cancer incidence of 9.8% (CI = 8.6 to 11.1) in men and 4.0% (CI = 3.3–4.9) in women with microcytic anaemia (common in persons with IDA), [42] which is comparable to our findings.

Previous research has rarely focused on the risk of overall cancer and specific cancer types in persons with AI in a general practice population [19]. A single and small-scale study showed that 23% of persons with AI had underlying malignancy, [19] which is higher than our findings. However, this may be due to different lengths of follow-up.

Previous findings on cancer risk and characteristics in persons with anaemia/IDA in general practice are in line with our findings [10, 11, 13]. Furthermore, the highest cancer risk occurred in CIIDA and AI, which are both characterized by an underlying inflammation. Inflammatory markers are associated with an increased cancer risk, and the risk increases as the level of inflammation rises [20]. Thus, the combination of inflammation and anaemia should increase the clinical alertness of underlying cancer. Moreover, across all investigated anaemia types, we found that persons with comorbidities were less likely to be diagnosed with cancer compared to persons without comorbidity. A reasonable explanation could be that patients with comorbidities may already have a reasonable explanation for the anaemia (i.e. comorbidity associated with anaemia, such as kidney disease and rheumatologic disease) [17, 43]. However, research seems needed on the cancer risk in persons with anaemia having certain comorbidities associated with an increased cancer risk, such as diabetes, [44‐46] cardiovascular disease, [45, 47] and inflammatory bowel disease [48].

Further, we found that women had a lower cancer risk compared to men. This could reflect that anaemia is a benign sign in premenopausal women due to e.g. menstrual bleeding. However, in a sub-analysis on the cancer risk in persons aged 50–90 years, the cancer risk in women compared to men increased only marginally (HR 0.86 vs. HR 0.84). This may reflect the overall increased cancer incidence in men compared to women, which is well-established and has been seen in the past decades in the Nordic countries [40].