Influence of inflammatory polyarthritis on cancer incidence and survival: Results from a community-based prospective study

Abstract

Objective

To investigate whether the incidence of cancer is increased and whether the rate of cancer survival is reduced in patients following the onset of inflammatory polyarthritis.

Methods

Between 1990 and 1999, we recruited 2,105 patients to a large primary care–based register of new-onset inflammatory polyarthritis. Subsequent cancers were ascertained by linkage to hospital and death records and were confirmed by the regional cancer register. Cancer incidence, both all-site and site-specific, was compared with regional rates, adjusting for age, sex, and calendar year. Overall cancer survival, adjusted for site, was compared with regional data using Kaplan-Meier curves and Cox regression.

Results

There were 123 incident cases of cancer in the cohort of patients with inflammatory polyarthritis. The overall incidence of cancer among this cohort of patients with inflammatory polyarthritis was not increased compared with that in the regional population. Among cancers of all major organ systems, only the incidence of hematopoietic cancers (including lymphoma) was increased. Five-year cancer survival was reduced in patients with inflammatory polyarthritis compared with patients without inflammatory polyarthritis. After adjusting for diagnosis, age, sex, and tumor type, mortality in patients with inflammatory polyarthritis and cancer was significantly increased (hazard ratio 1.4, 95% confidence interval 1.1–1.7).

Conclusion

This is the first investigation of overall cancer survival in patients with inflammatory polyarthritis. Compared with an increased incidence of cancer, reduced cancer survival might be a greater contributor to the increased cancer mortality observed in some rheumatoid arthritis populations.

It is well established that patients with rheumatoid arthritis (RA) have increased mortality compared with the general population. Typically, most studies show an increase of ∼2-fold (1–6). This doubling in mortality does not seem to be explained by the rheumatoid disease process in itself; it is more likely that an increased occurrence of comorbid conditions, such as cardiovascular disease, is responsible.

Several studies have also provided data on cancer mortality, with most showing only a modest increased risk in RA patients when compared with the general population (1, 2, 4, 7, 8). However, 2 studies demonstrated a substantially increased risk of cancer death, almost 2.5-fold that of the general population (6, 9). In one study (6), the frequency of disease-modifying antirheumatic drug (DMARD) use was particularly high (given the calendar period of recruitment), suggesting that RA was more severe in the patients being studied. In the other study (9), cancer mortality in a regionally derived Norwegian population of patients with RA was compared with national population data. Thus, the increased cancer mortality risk in that study may be explained by regional differences. Nonetheless, the data are consistent with an increase in cancer-related mortality in patients with RA.

There are 2 possible explanations for this increased mortality. First, the incidence of cancer in RA cohorts may actually be increased in comparison with that in the general population. Alternatively, patients with RA in whom cancer develops may have a higher case fatality rate when compared with unselected cancer patients.

With regard to the former possibility, most studies have demonstrated an increased incidence of cancer among patients with RA compared with the incidence in the general population (10–16). Again, those cohorts with the highest increases in cancer rates represent populations of patients in whom RA is particularly severe (12), such as those in which the level of DMARD use is high (11, 13, 14). This increased rate varies by cancer site. Specifically, patients with RA are at an increased risk for the development of lymphoma (16–19) and lung cancer (14, 15, 18, 20, 21). Conversely, colorectal and digestive tract cancers seem to occur less often in such cohorts (10, 15, 18, 21).

Investigations of the long-term outcome following the onset of RA in patients attending specialist clinics are subject to referral and selection bias. One strategy to overcome this is to follow up all patients in a target population with new-onset inflammatory polyarthritis (IP). Over time, an increasing proportion of new-onset IP evolves into RA, diagnosed according to the criteria of the American College of Rheumatology (ACR; formerly, the American Rheumatism Association) (22, 23). Thus, it is relevant in quantifying long-term risks of specific outcomes to ascertain the incidence in the entire IP cohort as well as that in patients who ultimately can be classified as having RA. We recently demonstrated that there is an increased risk of lymphoma in a primary care–derived sample of patients with new-onset IP (17). What is less clear is whether unselected patients with IP are also at an increased risk of either all-site or other site-specific cancers (excluding lymphoma).

No study has yet investigated whether patients with RA or IP in whom cancer develops have a reduced rate of survival. Here, we report results of a prospective study, the aim of which was to assess the risk of cancer, both overall and site-specific, in an unselected cohort of patients with IP. We also examined whether survival decreases following cancer onset, as well as the role of disease-related factors in predicting survival.

PATIENTS AND METHODS

Summary of design.

A prospective study was conducted following up a large primary care–based incident cohort of patients with IP. The risk of cancer was estimated by comparing the incidence of cancer (site-specific) in the IP cohort with that in the local population. Differences in survival were then compared between patients with cancer in the IP group and patients with cancer in the local population. Finally, the possible role of clinical and other factors in predicting mortality in patients with cancer within the IP cohort was assessed.

Subjects.

The IP cohort was derived from the primary care–based Norfolk Arthritis Register (NOAR), based in Norwich, UK. Methods for recruiting patients into NOAR are described in detail elsewhere (24). Briefly, from 1990 onward, patients ages 16 years or older who had synovitis of 2 or more joints lasting for at least 4 weeks were referred to NOAR by their primary care physician. Patients underwent a structured interview and physical examination at baseline, which were conducted by a trained research nurse. Additionally, serum samples were obtained for the determination of rheumatoid factor (RF) status. This analysis comprises all patients recruited between January 1, 1990 and December 31, 1999.

Rheumatologic followup.

Patients were followed up annually by a research nurse. At each followup visit, details on smoking exposure and all prescribed medications were obtained. Patients completed the Stanford Health Assessment Questionnaire (HAQ) (25), and RF status and smoking exposure were recorded. Radiographs of the hands and feet were obtained and scored for the presence of erosions, as described elsewhere (26). Patients were excluded if, during followup, a physician diagnosed a condition other than RA, psoriatic arthritis, or postviral arthritis as being the cause of peripheral joint IP. The revised ACR criteria for the classification of RA (22) were applied annually to the remaining members of the cohort, and patients were categorized as having RA if they satisfied the criteria cumulatively over the first 5 years of followup (27). The remaining patients were categorized as having undifferentiated IP.

Cancer ascertainment in the IP cohort.

Data on all cases of cancer occurring since January 1, 1995 were obtained by linkage of the cohort with the electronic records system of the region's only major hospital, which provided data regarding date and cause of all inpatient admissions (including day care) for all patients. As a test of the completeness of ascertainment, incident cancers were sought using 2 other approaches. First, all NOAR patients were asked during each annual followup whether they had experienced any serious comorbidity, including cancer, in the previous 12 months. Second, NOAR patients were linked to the National Death Register, which, for all patients who died, provided data on the date and cause(s) of death (underlying and contributory) when cancer was mentioned on the death certificate.

Use of regional cancer register.

Data from the Eastern Cancer Registration and Information Centre (ECRIC), which handles the entire region covered by NOAR, were used for 3 purposes. First, the data were used to verify the cancers ascertained in the NOAR cohort. Second, they were used to provide the source of data on cancer survival. Third, these data were used to derive comparative incidence rates for the background population.

In the UK, information on all tumors is submitted to regional cancer registries such as ECRIC. Primary cases of cancer are those considered by the diagnosing physician to be distinct from previously registered cases of cancer. Secondary cancers are those that the physician reports to be closely linked to a previously reported cancer (e.g., when a cancer metastasizes into another site). Only cancers submitted to ECRIC as new cancers during the study period were considered for analysis in this study.

Anonymized individual-level data on all cancers occurring since 1995 were provided by ECRIC. These data also included NOAR registration numbers (when applicable), thus enabling discrimination between cases developing in NOAR patients from those developing in the general population. This data set consisted of an anonymized patient number, date of the earliest mention of a tumor according to ECRIC data (effectively, the date of diagnosis), the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision code of the tumor, the last date of known contact, and the status of the patient at that point in time. ECRIC also receives notification of the death of any patient registered as having a tumor, through the National Death Register. Therefore, the date of last known contact will represent either the date of a patient's death or the last date at which a physician had contact with the patient. These data were then used to determine cancer survival, which was calculated from the date of first mention to either the date of death or the end of the study period (defined as March 31, 2004).

For the purposes of analysis, only the first-ever primary cancer occurring during the at-risk period was considered. Patients with known preexisting cancers (i.e., cancers diagnosed before the start of followup) were not excluded, because data on such cancers were unavailable for the comparison group (see below). However, cases of secondary cancer (i.e., a cancer that had spread from another site) were excluded.

Cancer ascertainment in the local population.

ECRIC was also used to obtain regional cancer incidence rates of all-site cancers first reported since 1995. For the purposes of analysis and to ensure comparability with the IP cohort, only the first primary cancer reported during this time period was considered. Secondary cancers and cases of nonmalignant skin cancers and carcinoma in situ were excluded. These data were used to produce 10-year age (at cancer diagnosis), calendar-year, and sex-specific incidence rates for each cancer site of interest. In this case, the denominator was the age- and sex-stratified population living in Norfolk in that year (i.e., the number of persons, per year, at risk). Rates from 2002 were used in place of rates from 2003 and 2004, because data from those years were unavailable. As with the IP cohort, survival was calculated from the date of cancer diagnosis to the date of last contact.

Statistical analysis.

The followup period started 12 months after the onset of IP symptoms (to reduce the likelihood that the cancer predated the onset of arthritis) or the start of 1995 (because hospital linkage data were available only for the period starting January 1, 1995) and ended on the date of death, cancer diagnosis, emigration, or March 31, 2004 (the end of the study period).

The site-specific incidence of cancer per 10,000 person-years was calculated overall and then stratified by sex for both the NOAR cohort and the local population. Adjusted relative risks (RRs) were obtained by comparing the cancer incidence in the NOAR cohort with the age-, sex-, and calendar year–specific rates obtained in the local population (see above), and the 95% confidence intervals were calculated.

Kaplan-Meier curves were used to compare survival among patients with cancer in the 2 cohorts, after adjusting for differences in age, sex, and distribution of tumor types. Adjustment for the latter was undertaken because cancers at differing sites obviously have different mortality hazards. A Cox regression was then conducted to calculate the hazard ratio (HR), comparing the IP cohort with the background Norfolk population after adjusting for differences in age, sex, and tumor type.

Finally, for those patients within the IP cohort in whom cancer had developed, the roles of the mean HAQ score over the first 5 assessments, cumulative RA status after 5 years, presence of erosions, DMARD use, and smoking history in predicting survival were assessed univariately (adjusted for by age and sex), using a Poisson regression model. For the purpose of external validity, analyses were also undertaken with stratification for RA status.

RESULTS

From the start of 1990 until the end of 1999, 2,318 patients were recruited to NOAR. Of these, 158 patients subsequently received a diagnosis of another specific condition explaining their symptoms and were thus excluded. As mentioned above, arthritis in the remaining patients was considered, during the course of their illness, to be attributable to RA, psoriatic arthritis, or any other unspecified inflammatory arthritis, including postviral arthritis (regardless of whether a viral etiology had been proven). Seven patients were excluded due to missing data on the year of death, and a further 3 patients who were not followed up for at least 12 months after the onset of IP were also excluded. Linkage with the hospital electronic admissions system was possible for all but 45 patients who had died before the start of collection of these data. Demographic data on the remaining 2,105 patients (overall and by RA status, to assist clinical comparison of the NOAR cohort) are shown in Table 1. As anticipated, the patients classifiable as having RA were much more likely to have erosive disease and to have been prescribed DMARDs.

Table 1. Characteristics of the IP cohort, stratified according to cumulative RA status*

Characteristic	All patients (n = 2,105)	Non-RA (n = 868)†	RA (n = 1,237)†
* Except where indicated otherwise, values are the number (%). IP = inflammatory polyarthritis; DMARDs = disease-modifying antirheumatic drugs; HAQ = Health Assessment Questionnaire. † According to rheumatoid arthritis (RA) criteria applied cumulatively at fifth assessment. ‡ Based on results in 1,959 patients (n = 767 non-RA and n = 1,192 RA). § Based on results in 1,441 patients (n = 378 non-RA and n = 1,063 RA). ¶ Based on results in 1,154 patients (n = 453 non-RA and n = 701 RA).
Age at onset of IP, mean ± SD years	53.7 ± 16.1	50.2 ± 16.5	56.1 ± 15.3
Female sex	1,408 (67)	582 (67)	826 (67)
Rheumatoid factor positive‡	681 (35)	117 (15)	564 (47)
Ever use of DMARDs	986 (47)	228 (26)	758 (61)
Erosive joints§	695 (48)	71 (19)	624 (59)
HAQ score ≥1.0 at fifth anniversary¶	518 (45)	202 (45)	316 (45)

The total followup was 15,548 person-years, with a median per-patient followup of 8.4 years. During this period of time, 130 cases of cancer (excluding nonmelanotic skin cancers and carcinoma in situ) were identified using hospital record linkage. No additional cases were revealed using annual followup interviews, as described above. However, a review of the death certificate data revealed an additional 17 cases of cancer. All 147 cases of cancer had been captured by the regional cancer registry. However, 20 cases were excluded because tumor registration took place either before or within 12 months of the onset of symptomatic arthritis. Another 4 cases were of secondary cancer and were thus excluded from analysis. Therefore, 123 cases of cancer were considered in this analysis.

Site-specific cancer incidence rates per 10,000 person-years for the overall cohort and according to sex are shown in Table 2. Although some specific cancers occurred more frequently in women, cancers overall occurred almost twice as frequently in men. Neoplasms of the breast, male and female urinary/reproductive organs (vulva, vagina, cervix, isthmus, placenta, penis, testis, kidney, renal pelvis, urethra, and bladder), lung, and colon or rectum were the most common in the NOAR cohort overall. Lymphoma was the sixth most common cancer observed.

Table 2. Relative risks of inflammatory polyarthritis according to tumor site*

Site	NOAR cohort	Norfolk population	RR†	95% CI
* Values are the number of cancers (incidence per 10,000 person-years). The Norfolk population comprises the age- and sex-stratified population of persons with cancer (ages 16 years and older) living in Norfolk. NOAR = Norfolk Arthritis Register; RR = relative risk; 95% CI = 95% confidence interval; CNS = central nervous system. † Adjusted for age and sex. ‡ Excluding nonmelanoma skin cancers.
Oral/facial			0.5	0.1–3.7
Men	0 (0)	318 (1.3)
Women	1 (0.9)	161 (0.6)
Colorectal			0.9	0.5–1.4
Men	9 (18.2)	2,308 (9.1)
Women	7 (6.6)	2,026 (7.5)
Other digestive system			1.3	0.8–2.1
Men	11 (22.3)	1,801 (7.1)
Women	4 (3.8)	1,297 (4.8)
Lung, bronchus, trachea			1.1	0.7–1.8
Men	7 (14.1)	2,470 (9.7)
Women	10 (9.4)	1,283 (4.8)
Other respiratory system			0.9	0.1–6.4
Men	1 (2.0)	219 (0.9)
Women	0 (0)	69 (0.3)
Bone, skin, and tissue‡			1.0	0.4–2.5
Men	4 (8.1)	682 (2.7)
Women	1 (0.9)	578 (2.1)
Breast			0.8	0.6–1.3
Men	0 (0)	37 (0.1)	0.6	0.3–1.2
Women	22 (20.7)	4,855 (18.0)
Prostate, men	9 (18.2)	3,704 (14.6)
Urinary/reproductive organs			1.0	0.6–1.6
Men	7 (14.2)	1,541 (6.1)
Women	11 (10.4)	2,711 (10.1)
Brain and CNS			–	–
Men	0 (0)	413 (1.6)
Women	0 (0)	414 (1.5)
Hematopoietic			1.6	1.0–2.7
Men	7 (14.1)	1,330 (5.2)
Women	8 (7.5)	1,115 (4.1)
Carcinomatosis			1.0	0.4–2.6
Men	0 (0)	493 (1.9)
Women	4 (3.8)	558 (2.1)
Secondary tumor			1.4	0.5–3.7
Men	1 (2.0)	370 (1.5)
Women	3 (2.8)	310 (1.2)
All primary sites‡			0.9	0.7–1.1
Men	55 (111.4)	15,316 (60.3)
Women	68 (64.1)	15,067 (56.0)

These rates were then compared with those in the general Norfolk population (individuals ages 16 years and older), as seen in Table 2. Because there were differences in age and sex distribution between the NOAR cohort and the general population, the RRs were adjusted for age and sex. Across all sites, the incidence of cancer (excluding nonmelanoma skin cancers and carcinoma in situ) was not increased in the IP cohort compared with the general population. The risk of hematopoietic cancers was increased, which was expected given the association between RA and lymphoma (17). Although there was evidence of an increase in the risk of digestive (excluding colorectal) neoplasms and secondary tumors, these risks were not statistically significant.

Figure 1 shows the observed and predicted survival curves for each cohort. Survival in the NOAR (IP) cohort was clearly reduced compared with that in the Norfolk (general population) cohort, with evidence that the risk was constant over time. We therefore undertook a Cox proportional hazards regression analysis adjusting for age, sex, and cancer site. Cancer survival was obviously reduced with increasing age, with a 4% increase in mortality per year. Male patients had reduced cancer mortality (HR 0.92, 95% CI 0.89–0.95), possibly reflecting the differences in the site of cancer. Therefore, after adjustments were made for age, sex, and site, IP patients with cancer had a statistically significant increased risk of death compared with the general Norfolk population with cancer (HR 1.4, 95% CI 1.1–1.7) (see Figure 1). It would be interesting to investigate whether the reduction in survival was similar across all cancer sites, but the numbers of cancers at the individual sites was too small for a robust analysis. However, as examples, the increased risk of all digestive system cancers (i.e., colorectal and other digestive system cancers combined for increased power) and hematopoietic cancer was of an order of magnitude similar to that for the overall risk of cancer (Figure 1).

Figure 1. Observed and predicted survival curves for the Norfolk Arthritis Register (NOAR) and general population (Norfolk) cohorts. Cox regression was used to calculate the hazard ratio (HR), comparing the NOAR cohort with the Norfolk population, after adjusting for differences in age, sex, and tumor type. The lines for observed survival (Norfolk) and predicted survival (Norfolk) are completely overlapping. HR = hazard ratio; 95% CI = 95% confidence interval.

Finally, we investigated whether any aspects of disease severity or therapy were useful predictors of cancer survival in the 127 patients with cancer in the IP cohort (Table 3). In general, all of the markers of disease severity were associated with increased mortality, although these increases were modest, and the 95% CIs spanned unity. Ever use of DMARDs was not associated with any reduced survival; there were too few patients to undertake a detailed analysis of the contribution of specific DMARD regimens on survival. Cigarette smokers had worse survival, but this was not statistically significant.

Table 3. Risk of death in patients with IP and cancer, according to potential predictors*

Factor	RR	95% CI
* IP = inflammatory polyarthritis; RR = relative risk; 95% CI = 95% confidence interval; RA = rheumatoid arthritis; HAQ = Health Assessment Questionnaire; DMARD = disease-modifying antirheumatic drug.
Rheumatoid factor positivity	1.29	0.84–1.99
Erosions ever	1.05	0.64–1.71
Cumulative RA status positive at fifth assessment	1.13	0.71–1.79
Mean HAQ score >1 over first 5 assessments	1.14	0.95–1.37
DMARD ever use	0.97	0.55–1.71
Smoking ever	1.53	0.92–2.54

DISCUSSION

Although we observed no evidence of an increased risk of all-site cancer morbidity in patients with IP, there was the suggestion of an association between IP and some site-specific cancers. The major finding was that patients with recent-onset IP in whom cancer developed were subject to reduced cancer survival over 5 years when compared with cancer patients in the general population (even after adjusting for age, sex, and cancer site). This reduced survival was independent of differences in age, sex, and cancer site distributions.

This study has some limitations that need to be discussed. Subgrouping the cancers by site revealed a relatively low number of cancers in each organ, and, thus, the CIs for the site-specific RRs were wide. Similarly, although the study had sufficient power to identify a reduction in cancer survival overall, it was not possible to evaluate whether this effect was different between cancer sites.

We were restricted in our analysis to only those cancers ascertained after January 1, 1995, due the introduction of the linked record system. Given the lag phase of 12 months following onset adopted in this analysis, we excluded cancers developing in patients who were recruited between 1990 and 1993, before the 1995 start date. Given that the relevant exposure periods were also excluded, the person-years studied in this analysis are not a complete record of all the followup time for the entire cohort. It is unlikely that this would have induced any important bias in the risk estimates of cancer incidence. Nonetheless, we performed a subgroup analysis of the patients who were recruited after 1993. This subgroup analysis resulted in widened CIs, but the effects described were not materially altered. The overall cancer incidence was slightly higher in this subgroup (85.4/10,000 person-years; 95% CI 69.8–102.5) compared with the incidence in the cohort as a whole (79/10,000 person-years), and the RR was also slightly increased in this subgroup (RR 1.1 versus 0.9 in the entire cohort). This partially reduced period of followup also would not have affected the comparison of cancer survival.

This analysis was concerned only with new cancers that developed during the period of observation. However, data were available on previous cancer history in the NOAR cohort. These data showed that of the 123 patients in whom cancer developed during the followup period, only 9 (7.3%) had a record of another malignant condition predating the onset of IP. Because the aim of the study was to identify all new malignancies developing during the followup period, it was deemed unnecessary to exclude patients with evidence of preexisting malignancies. Furthermore, data on preexisting cancers were not available in the general population sample, and hence the same rules applied to both cohorts.

The analyses presented considered the number of patients with cancer, rather than the number of cancers, as the numerator. It is thus of interest to consider how many patients developed >1 cancer during the followup period. In fact, only 7 (5.7%) of the NOAR patients received the diagnosis of a second cancer during the followup period, compared with 3.3% of the general population. Hence, the overall effect of including all such “additional” cancers rather than taking a patient-specific approach would be a slight increase in the RRs obtained. In terms of cancer survival, it seems reasonable to suggest that the slightly greater number of second cancers in the NOAR cohort may account for only a small part of the 40% increase in mortality in patients who had both IP and cancer. Importantly, this point relates to additional primary cancers. As a rule, cancer recurrences are not registered as new cancers and thus do not constitute a part of any comparison data provided by cancer registries (although data on such cancers are recorded, they were not made available).

Because cancer ascertainment relied mainly on the use of admission data from a local hospital, it is possible that our results precluded those cases of cancer that occurred in patients attending other hospitals. However, the hospital from which these data were obtained is the only main hospital serving the study region; thus, it is unlikely that use of data from only this hospital greatly influenced our results. Further annual interview of the patients revealed no additional cases of cancer.

In this study, the disease cohort comprised patients with undifferentiated IP rather than individuals who met the ACR classification criteria for RA at baseline. We previously showed that application of the ACR criteria at baseline is unstable in patients with early arthritis (22, 28). These results therefore apply to a primary care–derived population with new-onset IP. The use of such patients results in the inclusion of those with generally less severe disease than that reported from hospital-derived cases, and thus the increased risk in the latter group may be even higher. Of interest, ∼60% of the patients with IP examined met the ACR criteria for RA during the first 5 years of followup. When analyses were repeated, restricting inclusion to this group, there was little overall difference in the results. The overall cancer incidence in the subgroup with RA was substantially increased: 99.2 cases/10,000 patient-years (95% CI 80.0–121.7) compared with that in the group as a whole, although after adjustment for age and sex the RRs were virtually identical in the RA subgroup (RR 1.1, 95% CI 0.9–1.3) and the group as a whole. The overall increase in mortality in the RA subgroup was also similar to that in the whole IP cohort (HR 1.3, 95% CI 1.0–1.8).

It is widely accepted that patients with RA are subject to an increased risk of death from a range of comorbid conditions other than cancer. As such, it is possible that the observed reduced survival was attributable to the increased mortality that is generally observed in such populations. Because cause-of-death data were available only for the IP cohort, it was not possible to fully explore this issue. However, within the IP cohort, 80% of the patients with cancer who subsequently died had cancer listed as the main cause of death. The next most common causes of death were respiratory disease (9%), cardiovascular (including cerebrovascular or arterial) disease (7%), and gastrointestinal disease (2%). One patient died as a result of Alzheimer's disease, and another died of RA. Remarkably, similar cause-of-death patterns were observed in a large Scottish record linkage study of patients with cancer (29). Therefore, it seems unlikely that the observed differences in cancer survival seen in patients with IP are the result of competing comorbidities.

Because patients with IP receive relatively regular medical attention, it is possible that they are subject to screening bias. Such screening bias would lead to an increased likelihood of a cancer being detected sooner, thus explaining any increased risk of indolent cancers (including some subtypes of lymphoma) influencing the survival estimates. However, there was no significant difference between the IP cohort and the general population cohort in the proportion of cancer cases diagnosed at death (2.2% versus 3.1%; T score 0.49, P = 0.62), as one might expect if such a screening bias were applicable. Furthermore, any such bias would actually artificially increase the survival of patients in the IP cohort, thus implying that our results actually underestimate the reduction in cancer survival in patients with arthritis.

This study has several strengths. The fact that it is the first study of cancer incidence in a primary care–derived cohort of patients with IP removes any possible bias resulting from the use of hospital- or clinic-based patients. Furthermore, this is the first-ever investigation of cancer survival in patients with IP or RA in general. The truly prospective nature of this study is a particular strength, providing us with virtually complete and highly reliable data. The electronic linkage was also virtually complete, with only 55 patients (2.5%) excluded from followup, as detailed above.

The risk estimates for site-specific cancers are consistent with the findings of similar studies. There was a slight increase in the incidence of lung cancer as well as a slight decreased risk of colorectal cancers, although the small numbers resulted in wide CIs. The risk of hematologic cancers was increased, as recently reported in relation to an increased risk of lymphoma (17). This has also been widely reported by other investigators (10, 15, 16, 18, 21, 30).

It remains uncertain why patients with IP experience reduced cancer survival compared with the general population. Possible explanations include a role of general ill health, which might also limit the aggressiveness of the cancer therapy adopted, and an increased likelihood of infection, although the latter was not an important factor in the cancer deaths in this cohort. Analyses of disease markers associated with mortality in patients with both IP and cancer suggest that high levels of RF and a history of smoking increase the risk of cancer-related mortality. Future research into identifying risk factors associated with poorer cancer survival in such patients is needed so as to recognize individuals at the greatest risk. Prior DMARD therapy did not appear to be of relevance, although it would be necessary to undertake a more detailed study, probably with larger numbers of subjects, to truly exclude the possible role of either specific DMARDs or factors such as combination therapy, dose, and duration of therapy.

A history of recent malignancy is accepted as precluding the prescription of anti–tumor necrosis factor therapy. However, given the findings regarding cancer recurrence risks in patients treated with such therapies (31), the role of IP itself on reducing survival needs to be taken into account. Current guidelines suggest that the risk of future malignant disease may outweigh the benefit of biologic treatments in patients in whom cancer was diagnosed up to 10 years previously (32).

In summary, the results of this study demonstrate that 5-year cancer survival in patients with IP is substantially reduced in comparison with that in the general population (even after adjusting for differences in age, sex, and cancer site), whereas the overall cancer incidence does not seem to be increased. As such, it would seem that the well-reported increased rate of cancer mortality observed in patients with IP (4) is most likely attributable to a reduction in cancer survival.

AUTHOR CONTRIBUTIONS

Dr. Silman had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Franklin, Symmons, Silman.

Acquisition of data. Franklin, Bunn, Symmons, Silman.

Analysis and interpretation of data. Franklin, Lunt, Symmons, Silman.

Manuscript preparation. Franklin, Lunt, Symmons, Silman.

Statistical analysis. Franklin, Lunt, Silman.

Acknowledgements

The substantial support of Professor David Scott and clinical colleagues at the Norfolk and Norwich University Hospital and the local primary care physicians is gratefully acknowledged. We are also grateful to Helen Coffey for undertaking the linkage to the hospital record systems and to the Eastern Cancer Registration and Information Centre for providing cancer data on the observed and comparison populations.