Objective To determine the risk of cancer before and after the diagnosis of motor neuron disease (MND), multiple sclerosis (MS) and Parkinson's disease (PD).
Methods Analysis of statistical database of linked statistical abstracts of hospital and mortality data in an area in southern England.
Results Only people with PD showed a significant difference in the overall incidence of cancer compared with controls (rate ratio (RR) 0.76, 95% CIs 0.70 to 0.82 before PD; RR 0.61, 0.53 to 0.70, after PD). RRs were close to 1 for cancer in patients after MND (0.98, 0.75 to 1.26) and after MS (0.96, 0.83 to 1.09). There were high rate ratios for malignant brain cancer (7.4, 2.4 to 17.5) and Hodgkin's lymphoma (5.3, 1.1 to 15.6) in patients diagnosed with MND after cancer. In people with MS, malignant brain cancer also showed an increased RR both before hospital admission with a diagnosis of MS (3.2, 1.1 to 7.6) and after (2.4, 1.2 to 4.5). In people with PD, several specific cancers showed significantly and substantially reduced RRs for cancer, notably smoking related cancers, including lung cancer (0.5, 0.4 to 0.7, before PD; 0.5, 0.4 to 0.8, after PD) but also cancers that are not strongly smoking related, including colon cancer (0.7, 0.6 to 0.9, before PD; 0.5, 0.4 to 0.8, after PD).
Conclusions People with MND, or MS, do not have an altered risk of cancer overall. There may sometimes be misdiagnosis between MND or MS and brain tumours. PD carries a reduced risk of cancer overall, of some smoking related cancers and of some cancers that are not smoking related.
- Motor neuron disease
- Multiple sclerosis
- parkinson's disease
Statistics from Altmetric.com
Disease association studies, to investigate whether diseases coexist in individuals more or less commonly than expected by chance, can provide insights into diseases of obscure aetiology. There is a rich case report literature on cancer, as a marker for a number of different pathological processes, in people with motor neuron disease (MND), multiple sclerosis (MS) and Parkinson's disease (PD). The epidemiological literature is more sparse.
The aetiology of MND, and amyotrophic lateral sclerosis in its commonest form, is poorly understood. A minority of cases are associated with single gene mutations, and although there may be several inherited susceptibility genes, the disease is best considered as a sporadic disorder.1 An association between MND and cancer has been postulated in the literature, largely based on case reports with relatively small numbers of cases.2 3 Some studies have reported remission of an apparently paraneoplastic form of MND on treatment of the underlying tumour4 5; and associated antineuronal antibodies (especially anti-Hu and anti-Yo antibodies) have been sought and found by some authors6 7 but not found by others.8
The aetiology of MS, an inflammatory, demyelinating disease of the central nervous system, is similarly thought to involve a ‘complex genetic risk profile’ but in typically younger people than MND and with clearer evidence about the role of environmental influences.9 The literature about MS and cancer has mainly focused on Hodgkin's lymphoma and the evidence that Epstein–Barr virus infection may be implicated in the aetiology both of it and of MS,10–12 and on breast cancer13 and brain tumours.14 Elevated cancer rates have also been suspected to result from immunosuppressive pharmacological therapy for MS.15
PD, in contrast, has been reported to be associated with significantly low rates of some cancers but also with significantly high rates of breast cancer and melanoma.16 Some of these associations have generated, or been taken to support, various hypotheses, including the possible neuroprotective effects of smoking (lower rates of smoking related cancers in PD), L-dopa treatment as a cause of malignant melanoma and genetic mutations as a common aetiological factor in both PD and cancer.16 17
Diseases can coexist in an individual by chance. We have used a large epidemiological database to study cancer in people with MND, MS or PD, with the aim of distinguishing between chance and true associations.
Materials and methods
Population and data
The Oxford Record Linkage Study (ORLS) includes brief statistical abstracts of records of all hospital admissions, including day cases, in UK National Health Service (NHS) hospitals, and all deaths in a defined population in southern England from 1963 to 1999.18 The hospital data were collected routinely in the NHS as the (former) Oxford Regional Health Authority's hospital discharge statistics. The death data derive from death certificates. The current programme of analysis of the data has been approved by the English NHS Central Office for Research Ethics Committees (reference number 04/Q2006/176). With changes to NHS information systems, data in the ORLS cannot be linked between the pre-1999 era and the present system of data collection (see Appendix for further details).
Construction of cohorts
The methods used to study people with each neurological condition were the same and will be described for MND. For people with MND, their first admission to hospital or day case care during the study period was identified. A control cohort was constructed by identifying the first admission of individual patients in the ORLS for a range of common orthopaedic, dental, ENT and other relatively minor disorders (details in table footnotes). This is based on a ‘reference’ group of conditions that has been used in a series of other studies of cancer in people with other non-malignant disease.19–21 We searched the dataset for any subsequent hospital admission or day case care for, or death from, each cancer in these cohorts. We used the reference cohort from within the ORLS dataset, rather than external population based rates, to calculate expected numbers of people with each cancer in the MND cohort. We did so because there is migration into and out of the Oxford region, without the ability to follow the medical records of those who move. This means that while absolute rates of cancer occurrence after MND cannot be calculated, relative rates, comparing the MND and reference cohort, can (see below).
We calculated rates of each cancer based on person years at risk. We took ‘date of entry’ into each cohort as the date of first admission for the exposure condition, or the reference condition, and we took ‘date of exit’ for each individual disease as the date of subsequent admission for cancer (if any occurred), death or 31 March 1999, whichever was the earliest.
In comparing the MND and reference cohorts, we first calculated rates of each subsequent cancer, standardised by age at entry (in 5 year age groups), sex, calendar year of first recorded admission, interval from study entry and district of residence, using the indirect method of standardisation and taking the combined MND and reference cohorts as the standard population. We applied the stratum specific rates from the standard population to the MND cohort and to the reference cohort, in order to obtain standardised rates for each cohort. We then calculated the ratio of the standardised rate of each cancer in the MND cohort relative to the rate of the same cancer in the reference cohort, with the 95% CI for the rate ratio, using the methods of Breslow and Day.22 In the reference cohort, we included all people in the dataset with the comparison conditions in each age group. We did this to maximise the numbers in each stratum in the reference cohort and thereby to maximise the statistical power of the study. Our purpose in standardising for age, year of admission and district of residence was to ensure that the exposure cohorts and the reference cohort were equivalent in these respects. We give figures for rate ratios and their 95% CIs at one decimal place except that, in analyses with all cancers combined, we show two decimal places. We then used the same procedures, but in reverse, to calculate rate ratios for MND (and MS and PD) after each cancer (see Appendix). Because we make multiple comparisons, we have given exact p values so that the reader may assess the level of significance of each comparison. In comparisons with an ‘expected’ value of 5 or less, we have quoted the exact test based on the binomial distribution.22
The age and sex distributions of the patients are shown in table 1. The mean period of follow-up from neurological disease to first occurrence of cancer was 3.1 years for MND, 6.9 years for MS and 3.2 years for PD; that from cancer to neurological disease was 3.4 years.
Motor neuron disease
The ORLS data (table 2) showed no significant overall association between cancer and MND. The rate ratio (RR) for cancer diagnosed after MND, relative to cancer in the control cohort, was 0.98 (95% CI 0.75 to 1.26). The RR for MND after cancer was 0.84 (95% CI 0.66 to 1.07). Within this group the RRs were significantly high for malignant brain tumours (RR 7.4, 95% CI 2.4 to 17.5) and Hodgkin's lymphoma (5.3, 1.1 to 15.6) prior to MND diagnosis but these associations disappeared if patients who were diagnosed with these conditions within a year of each other were excluded (see footnote to table 2 for RRs). RRs were high, in patients diagnosed with MND first, for cancer of the salivary gland (two observed cases, 0.2 expected; RR 9.5, 1.1 to 35.0) and low for prostate cancer (no observed cases, 6.1 expected, p=0.007) (table 2).
There was no significant association between MS and cancer overall, irrespective of whether MS was diagnosed before cancer (RR 0.96, 95% CI 0.83 to 1.09) or after cancer (0.88, 0.70 to 1.09). Malignant brain cancer showed an increased RR both after an admission for MS (2.4, 1.2 to 4.5) and before it (3.2, 1.1 to 7.6). The RRs for kidney cancer (2.1, 1.0 to 4.0) and bone cancer (4.0, 1.5 to 8.9) after MS were significantly elevated and that for non-melanoma skin cancer after MS was reduced (0.4, 0.2 to 0.9) (table 3).
There was a highly significant reduction (p<0.001) in the risk of cancer in patients with PD, both after the first recorded admission with PD (RR 0.61, 95% CI 0.53 to 0.70) and before it (0.76, 0.70 to 0.82). Several specific cancers showed markedly low RRs, most strikingly lung cancer (0.5, 0.4 to 0.8, after PD; 0.5, 0.4 to 0.7, before PD), bladder cancer (0.5, 0.3 to 0.9, after PD; 0.7, 0.6 to 0.9, after PD) and colon cancer (0.5, 0.4 to 0.8, after PD; 0.7, 0.6 to 0.9, before PD). Cancers of the oropharynx, stomach and prostate, non-melanoma skin cancer and non-Hodgkin's lymphoma were significantly low after PD; cancers of the oesophagus, rectum, pancreas and cervix, and malignant melanoma were significantly low before PD; and breast cancer was borderline significantly low both before and after admission for PD (all shown in table 4).
Strengths and weaknesses of the study
The strengths of using the ORLS dataset are that it is large in scale and it covers many years of ‘follow-up’ after a first diagnosis with the neurological diseases studied. The data were used to calculate ‘expected’ numbers of cases of diseases that occurred in combination, and therefore to assess the statistical significance of disease combinations. The use of a single population dataset to study the cancer profiles of different neurological diseases allows comparisons to be made between them: for example, the cancer profile of PD was very different from that of MND or MS. Moreover, measuring separately the numbers of cancer cases occurring before MND/MS/PD and the numbers occurring after MND/MS/PD provides some information about the temporal (and hence potentially the causal) nature of any associations found. For example, if a drug used to treat MS were thought to cause a particular cancer type, one might expect to see an excess of cases of this cancer being diagnosed after MS, but not before it. This is a particular strength of our methodology compared with past epidemiological studies, which generally investigate only one temporal direction of association or else do not distinguish between the two in data collection.10 15
The data also have some limitations. Data collection was limited to those in receipt of hospital care, although we think that the majority of people with MND, MS and most cancers, and quite a high percentage of people with PD, will have been admitted to hospital or seen as a day case at least once. Our data identified the first recorded hospital episode of neurological disease before cancer (and vice versa) but we cannot know whether the neurological diagnosis was invariably made before the diagnosis of cancer. There is, no doubt, some misclassification of some patients in the study, in the temporal sequence of disease occurrence. Accordingly, we have greater confidence in findings on disease associations that are similar regardless of which disease is recorded as having been hospitalised first. The data are confined to patients who remain within the area, and who are admitted to the hospitals, covered by ORLS data collection. These factors are part of our reasoning for including a comparison cohort from the same dataset and for tight ‘matching’, through stratified analysis, for district of treatment and for year of first recorded diagnosis as well as for age and sex. There are no data on lifestyle factors, notably none on smoking, which makes it impossible to explore some of the proposed mechanisms for cancer associations. Current privacy regulations preclude accessing original clinical records and checking on whether (for example) diagnoses of cancer were histologically confirmed. We have had to accept the coded diagnoses on the statistical summary record.
In studying a large number of associations between diseases, one must consider the effect of making multiple comparisons. For instance, with a level of significance of p<0.05, one expects one ‘significant’ result at this level for every 20 comparisons made by the play of chance alone. For this reason, we have given exact p values so that the reader may assess the level of significance of each comparison. Unexpected significant associations, especially with low numbers of observed cases, such as that between MND and cancer of the salivary gland (two observed cases), may be explained by chance. However, even with the number of comparisons made, associations where p<0.001 (eg, the low levels of cancer in people with PD) are very unlikely to be the result of chance alone.
Motor neuron disease
No significant association was found between MND and cancer overall. Despite the case reports implicating various cancers in MND, the overall lack of association between MND and cancer in our study, whether cancer was recorded before or after MND, accords with the limited epidemiological evidence to date.23 24 In the largest study thus far, Freedman et al used data from US cancer registries along with death certificates to ascertain cases of MND and showed no increase in MND occurrence in 1.9 million US cancer survivors (their standardised mortality ratio was 1.0 (95% CI 0.9 to 1.1).24 They also reported no excess or deficit of any specific cancers apart from an excess of melanoma, for which chance (due to multiple comparisons) or ascertainment bias were cited as potential explanations.
An association between breast cancer and MND in female patients has been noted in several case series2 3 25 but, as breast cancer is common, any association must be shown to be more than coincidence. Sadot et al reported seven female patients from their MND clinic with breast cancer, and concluded that there was an overrepresentation (23% compared with the normal female population of 12% cumulative lifetime incidence).3 As in previous studies,25 an upper motor neuron predominant form of amyotrophic lateral sclerosis was noted in the patients. The lack of any association in our data is in accordance with another large epidemiological study.24 However, despite the overall size of our study, it is nonetheless small and lacking in power in respect of individual site specific cancers. A paraneoplastic ‘MND-like disorder’ is likely to be a very rare and distinct entity from the sporadic disorder of MND, supported by the absence of both distinguishing clinical features in MND patients with cancer26 and common paraneoplastic antibodies in MND patient sera.8 However, the majority of neurological paraneoplastic disorders do precede the detection of the underlying tumour,27 which may be undetectable at subsequent autopsy if the patient succumbs rapidly. While this phenomenon is unlikely to confound results in relatively slowly progressive disorders such as MS and PD, the typically rapid progression of MND (2–4 years median survival from symptom onset) raises the potential for some under—ascertainment of underlying malignancies in this disorder.
Some apparent MND–cancer associations were demonstrated by the ORLS data. Most striking was that between malignant brain cancer and the subsequent development of MND. One plausible explanation might be the initial misdiagnosis of brain cancer in patients presenting with an apparent bulbar palsy. Diagnostic confusion between MND and other CNS cancers has been reported,28 with even ‘distant’ brain cancers occasionally known to yield bulbar palsies,29 but usually in the reverse situation—that is, MND misdiagnosed initially. The association between MND and a subsequent diagnosis of Hodgkin's lymphoma may also be attributable to misdiagnosis, or chance (only three cases observed, 0.6 expected), although a known mimic of lower motor neuron predominant MND (namely autoimmune multifocal motor neuropathy) has been reported in a case of B cell lymphoma.30 There appears to be a highly significant reduction in the incidence of prostate cancer in patients diagnosed with MND (RR 0; 95% CI 0 to 0.45, p=0.007) in our data, which is not reported in the literature. This may simply reflect the rapidity with which MND progresses, preventing the symptomatic manifestations of this common though often indolent cancer.
The absence of a significant association between MS and cancer, overall, accords with the epidemiological literature. A record linkage study, combining data from the Danish MS and cancer registers, found a non-significant decrease in cancer in patients with MS compared with the general population, with a standardised incidence ratio of 0.94 (95% CI 0.89 to 1.00).10 Our rate ratio—a comparable measure—was 0.96 (0.83 to 1.09).
The apparently increased incidence in our data of kidney and bone cancer after MS, and malignant brain cancer before and after MS, has not been reported in other epidemiological studies. However, case reports of MS and subsequent brain cancer have been published,14 31 with the suggestion of gliomatous transformation of demyelinating lesions. Reports of MS mimicking brain tumours clinically and radiologically could potentially explain the reverse association of brain cancer before MS,32 33 as could a brain tumour associated with T2 hyperintensity on MRI mimicking MS. The observed reduction in non-melanoma skin cancers in patients diagnosed with MS may be a reflection of the well established increased incidence of MS at higher latitudes where exposure to sunlight is diminished, and is a finding that has been reported from the ORLS previously.34
The relationship between MS and Hodgkin's lymphoma has been studied closely since Epstein–Barr virus, a closely associated agent in Hodgkin's (and Burkitt's) lymphoma, has been hypothesised as a risk factor for MS.35 The possibility of an association between MS and Hodgkin's lymphoma has received some support from case reports,35 a small Italian case control study12 and a study showing familial clustering.11 However, an association between MS and Hodgkin's lymphoma was not found in the large Danish record linkage study10 and, albeit that our numbers were small, there was no hint of an association in our study.
It has been suggested that an association between MS and subsequent breast cancer, found in some small case series, may result from immunosuppressive treatment for MS.13 15 Using data from European MS and cancer registries, Lebrun et al examined the cancer and pharmacological therapy profiles in 7418 MS patients and found a significant elevation in cancer risk (particularly for breast cancer) in patients taking immunosuppressive drugs to treat their MS.15 The authors commented on the carcinogenic potential of azathioprine and cyclophosphamide in MS and also suggested that the use of non-steroidal anti-inflammatory drugs in some patients may also have an effect on the cancer profiles by lowering the risk of many cancers, including breast cancer. Our data showed no significant increase in the incidence of breast cancer either after MS or before it. However, the RR for breast cancer after MS (1.20, 95% CI 0.9 to 1.6, p=0.12), while non-significant, nevertheless agrees closely with the value provided by the Danish record linkage study (standardised incidence ratio 1.21, 95% CI 1.05 to 1.39).10 Moreover, the RR for breast cancer before MS does not show this upward trend (RR 0.8, 95% CI 0.5 to 1.2), also in accordance with a mechanism involving drugs for MS increasing the rates of breast cancer. No data were collected in our study regarding the use of immunomodulatory or immunosuppressive drugs, and many of them were used only in a limited capacity in the timescale of our study, so we are unable to comment further on the potential role of these drugs in this context. Nevertheless, vigilance must be maintained, not least with the more widespread use of monoclonal antibody therapies, one of which (natalizumab) has been the subject of anecdotal reports of association with melanoma.36 Associations between two diseases, where one is a consequence of therapy for the other, will only be found in populations and at times where and when the therapy is used, and therefore ongoing study of these associations and the roles of new disease modifying therapies is required.
Our data showed a low incidence of cancer in PD. RRs were low for several smoking related cancers such as lung, bladder and oesophageal cancer, and also for some cancers that are not strongly smoking related, such as those of the colon and rectum. There is good evidence that colorectal cancer is associated with diet—for example, it is elevated in association with high protein and low fibre diet—and it is possible that people with PD tend to have had a ‘healthy lifestyle’ in respect not only of smoking but also other factors. A large Danish record linkage cohort study37 reported a borderline low standardised incidence ratio for colon cancer (RR 0.84; 95% CI 0.7 to 1.0) and showed low cancer risks at a number of cancer sites that are not smoking related.
Our results for smoking related cancers accord with a large Danish record linkage study and an American cohort study.37 38 It is conceivable that there is a higher than average level of smoking by people in our reference cohort (as stated above, we do not have smoking histories to judge) and, in theory, this could account for low levels of cancer in our neurological cohorts. This seems an unlikely explanation for the low levels of cancer in PD because the finding is specific for PD and not found in the MND or MS cohorts. A prevailing theory is that tobacco smoke, although it promotes some cancers, may also possess components that diminish the neurotoxicity involved in triggering PD.37 Prospective studies have demonstrated a significantly reduced incidence of PD among current and past smokers compared with those who have never smoked, and a dose–response relationship between PD incidence and time since quitting smoking.39 Twin studies have also shown that the risk of PD in both monozygotic and dizygotic twins is inversely associated in a dose dependent fashion with smoking.40 Nicotine seems the likely candidate to be the neuroprotectant, since it stimulates dopaminergic release and also protects against neuronal insults in experimental models.41 Genetic factors have also been hypothesised to play a role in the reduction in cancer incidence in PD patients, with mutations in parkin and other genes implicated in the tumorigenesis of several cancers.16
Previously reported evidence for an increased incidence of breast cancer and strong evidence for an increased incidence of malignant melanoma in PD were not reproduced in this study although the small number of expected cases of melanoma herein prevents strong conclusions.16 37 38 The increased incidence of breast cancer has been variously attributed to detection bias (although this would not explain reduced levels of most other cancers), elevated nocturnal prolactin in PD patients and increased oestrogen levels.16 The use of L-dopa to treat PD has been implicated as the agent linking PD with malignant melanoma since it is a substrate for the enzyme tyrosine hydroxylase which ultimately converts it to melanin and is overexpressed in melanoma tumour cells.16 Epidemiological evidence for this mechanism is lacking however.42
Record linkage datasets, based on routinely collected administrative data, provide useful tools for scanning for disease associations. Their findings should generally be regarded as suggestive rather than definitive. Unlike case control and cohort studies based on personal interviews with individual patients, routine data lack scope for identifying information about lifestyle, treatment and other potential causes and confounders. However, studies based on personal interviews, to ascertain the co-occurrence of two diseases both of which are uncommon, may be prohibitively expensive.
By contrast with most case series, population based, record linkage datasets can be used to identify ‘expected’ values for the co-occurrence of any two diseases. Our data showed no significant overall association between MND and cancer, or between MS and cancer. The apparent association of malignant brain cancers with both MND and MS is likely to be explained by misdiagnosis in the early stages of illness. Our data add further weight to the evidence that, in PD patients, there is a significantly low risk of cancers, including some cancers that are unrelated to smoking as well as those that are smoking related.
Over many years, the linked data files were built by Leicester Gill, Glenys Bettley and Myfanwy Griffith. The Unit of Health-Care Epidemiology is funded by the English NIHR Co-ordinating Centre for Research Capacity Development to analyse the linked data. MRT is supported through the MRC/MNDA Lady Edith Wolfson Clinician Scientist Fellowship.
Data collection covered part of one health district from 1963 (population 350 000), two districts from 1966 (population 850 000), six districts from 1975 (population 1.9 million) and all eight districts of the region from 1987 to 31 March 1999 (population 2.5 million). With the agreement of the Region's Data Protection Steering Group, the data for each individual were linked together routinely, as records accrued, as part of the region's health information system. The data are now anonymised and archived.
In analyses where we studied MND or the other neurological conditions first (the ‘exposure cohort’) and then each cancer (the ‘outcome’), we excluded individuals for whom the cancer was recorded before, or at the first admission for, MND or reference condition. Exclusion of ‘simultaneous’ cases avoided the potential bias of identifying people for the MND cohort (or reference cohort) simply because they had an admission for cancer. We then repeated the analysis for the next cancer until we had analysed the data for every cancer in turn. We then analysed each cancer in turn after MS and then each in turn after PD.
When we studied MND after cancer, we reversed the method as follows. Each individual ‘exposure’ cohort was now a cohort of people with each individual cancer and we ‘followed-up’ the individuals in each individual cancer cohort, and in the corresponding reference cohort, for the subsequent occurrence of MND. Follow-up was censored, on the same principles as previously described (see Method), by the date of subsequent occurrence of MND (if it occurred), death or the end of the study period. In studying MND after cancer, we excluded individuals for whom MND was recorded in the same admission as, or in an admission before, the cancer. We then repeated the analysis for each successive cancer in turn and repeated the analysis for each successive cancer followed by MS and then for each successive cancer followed by PD. The diagnoses of MND, the reference conditions, and each cancer were accepted as those coded on the hospital record abstract. Privacy regulations preclude access to the full records to study the criteria on which the diagnoses were based.
The flow chart and equations used for calculating the RRs and statistical inferences based on Breslow and Day22 are as follows:
Calculate the RR using equation 3.8.
Calculate the binomial parameters πL and πU using equation 3.9.
Substitute πL and πU into equation 3.6 to calculate the CIs for the rate ratio from (1).
Use equation 3.7 to calculate χ2 statistic and the appropriate SAS function (probchi(…)) to calculate the p value.
We grouped the conditions in the reference cohorts into four main groups. We show the numbers of cases in each group:
Gastro-inguinal hernia group (n=186 879): inguinal hernia (n=79 339), haemorrhoids (n=29 899), appendicectomy (n=77 641).
Lumps and bumps group (n=52 451): bunions (n=4617), sebaceous cyst (n=20 283), nasal polyp (n=27 551).
Trauma and orthopaedic group (n=213 316): knee internal derangement (n=26 394), dislocations, sprains and strains (n=23 066), superficial injury and contusion (n=29 965), fractures (n=98 046), hip replacement (n=26 495), knee replacement (n=9350).
‘Other’ group (n=298 854): upper respiratory tract infections (n=79 641), otitis media (n=68 573), diseases of nails (n=14 241), disorders of teeth (n=74 968), squint (n=18 040), varicose veins (n=43 391).
Funding English NIHR Co-ordinating Centre for Research Capacity Development.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.