Discussion
This large, population-based study of 1713 people with incident MS in UK primary care provides important data on the clinical characteristics of this patient population including prognostic factors for mortality. We observed that MS patients suffer from various conditions, with infections and depression the most common, but we also identified several other important predictors for all-cause mortality among MS patients, including smoking, alcohol abuse, multiple symptoms at first diagnosis, pneumonia and influenza, urinary tract infection, heart disease, and cancer.
Others have similarly found that people with multiple comorbidities had an increased risk of mortality [
16], and that cardiovascular disease and respiratory infections were associated with higher mortality in MS patients [
17‐
19]. A history of urinary tract infections was also found to be associated with decreased survival in these data; however, these observations are likely to reflect MS severity rather than a causal association. Sex was not associated with mortality in this study. The effect of sex has been shown to vary greatly across studies, over time, and between countries [
16‐
18].
An important objective of this study was to explore whether risk factors and comorbidities for MS interact with the underlying autoimmune process, or the drugs used to treat it, and the extent that these may influence mortality. As expected, we found that people with MS were at high risk of infections of all kinds possibly in part due to the use of immunosuppressant drug treatment. The strong association between pneumonia and risk of mortality could reflect the high risk of pneumonia infection in the MS population in concert with the high risk of death among those who develop pneumonia. We also found that patients who received MS treatments had a reduced risk of death; however, we cannot separate the effects of the drug from the effects of the comorbidities.
Patients who had infections and chronic comorbidities were often found to die of that underlying disease. Among the 29 MS patients who died and who had a history of pneumonia, the pneumonia was considered to be at least one of the causes of death in 22 (76 %) instances. Note that cause of death was not available for all patients who died and pneumonia may have been the cause of death for additional MS patients. This supports the notion that if patients have MS and develop pneumonia, the risk of death is high and perhaps higher than in the general population. Cancer was the known cause of death for 21 (81 %) of 26 MS patients who ever had a diagnosis of cancer in their record and subsequently died. Among the 18 MS patients who had a history of heart disease and died, 13 (72 %) died of cardiovascular disease. In all, 11 of the 115 MS patients who died had MS as the only listed cause of death.
We found that patients with multiple symptoms at MS onset had an increased risk of death compared with those with only sensory symptoms, among cases whose diagnosis was confirmed from their original clinical records during 1993–2006. In a large Finnish cohort of MS patients, Sumelahti et al. [
18] reported that optic neuritis or other sensory symptoms at presentation were associated with favourable survival. To the best of our knowledge, the association with multiple symptoms found in this study is new and should be investigated further.
Patients in this study with PPMS compared to RRMS were at increased risk for mortality (adjusted HR 2.2), although it was borderline significant. There were too few people with SPMS (adjusted HR 1.7) to obtain a reliable risk estimate. Compared with other existing studies [
16‐
19], our follow-up was relatively short which could explain the small number of cases, though, since these forms of MS are more severe the higher risk was expected. The average length of follow-up from first MS diagnosis was 7.7 years in our study. The highest HR, however, was found among people with unknown MS type. First, this result was based on a small number of patients and is thus not stable. Second, we were less likely to obtain original clinical records for patients who had died, so this group of MS patients is not a random sample of all MS patients but may represent patients with more severe types of MS, which could explain the strong association with risk of mortality.
We found that more patients died from cancer than from heart disease in this study. Since most MS patients in this study were female, this finding mainly reflects cause of death in women where CVD is not a major cause until after menopause. It is noteworthy that more men died from CVD than from cancer in this study. Since the mean age of death in this study was around 57, this finding is consistent with the leading causes of deaths in White women of the same age [
20]. The limited follow-up time in this study precluded following people until later age when mortality rates are higher. An alternative explanation is that MS patients could have better control of vascular risk factors compared to the general population. This could be the result of the close medical monitoring received by these patients and the fact that, confronted with a chronic and incurable disease, physicians will strive to correct those comorbidities amenable to treatment in an effort to decrease the total morbidity burden of their patients. Finally, it is also possible that the autoimmune process underlying MS, or the immunomodulatory agents used to treat it, or a combination of both, results in a slower progression of the atherosclerotic process that leads to heart disease.
Another relevant finding of the present study is the association between the infectious processes and mortality in MS patients. It is indeed feasible that this is a spurious association, possibly due to confounding generated by un- or incompletely measured factors associated with both infection occurrence and death. One obvious candidate for this inadequately controlled confounding is treatment for MS, for which we did not have complete information. Therapeutic strategies to confront MS involve modulating the immune system and could thus be related to the incidence and prevalence of infections via decreased immunity, and to mortality via either adverse reactions or disease severity. However, the observation that infections represent a common cause of death in other chronic conditions like Parkinson’s disease [
21] suggests that the described statistical association between infection and mortality could also represent a true biological phenomenon. One causal structure that could account for the described association involves the relationship between infections and treatment in terms of mediation rather than confounding: MS immunomodulatory therapies predispose MS patients to suffer infections, which in turn cause death.
Strengths of our study include the large sample size and high quality database. In addition, our findings are generalizable to the UK population as a whole because of the representative population-based nature of the CPRD. Additionally, we were able to validate MS diagnoses by accessing original patient clinical records for a large proportion of cases, which, along with linkage to death registry data, enabled additional clinical information to be obtained. Finally, we used age as the time-scale in the Cox proportional regression models instead of time-on-study, the scale used in most previously reported MS studies, to better estimate the effects of predictive factors for death controlling for age. Moreover, because patients may have developed MS many months or years prior to their first MS diagnosis, if we had estimated mortality where we followed patients from the first MS diagnosis, left truncation would have been present in this study. Using age as the time scale accounted for the left truncation issue [
22] and is more appropriate in this study.
We were, however, unable to validate the MS diagnosis via original records for all patients; thus, it is likely that some misclassification occurred. Any misclassification would likely have been random, non-differential, and any effects on the HRs would likely have been small and biased towards the null. We were also unable to describe the patient population regarding vitamin D status or ethnicity as this information is not systematically recorded by GPs. While it is thought that vitamin D deficiency is associated with the risk of cancer [
23] and cardiovascular diseases [
24], the magnitude of these associations is not strong; thus, the absence of information is unlikely to materially impact the results of this study. Recent population-based studies have also reported significant differences in the incidence of MS across ethnic groups. Since the vast majority of people in the UK are Caucasian it is unlikely that the inability to control for ethnicity would materially alter the results. Additionally, we were unable to evaluate the influence of the different MS treatments, particularly interferon beta which has been shown to slow disease progression [
25,
26] and reduce all-cause mortality [
27,
28] in patients with RRMS. Because in the UK, it is mostly prescribed in secondary care and not always captured in GP records, there were very few MS patients in this study who had records for interferon beta. This may have affected the estimates of all-cause mortality among RRMS patients in our study, yet is unlikely to have affected the HRs associated with the observed predictive factors for all-cause mortality. Finally, we did not have information on socioeconomic status (SES) for patients in this study, so we could not evaluate its effect on mortality. Since all people in the UK are covered by the National Health Service, access to medical care due to SES should not have had a major effect on MS care. However, we still cannot rule out the possibility that our results are confounded by SES.
In conclusion, our population-based study suggests that several factors, some of which are modifiable, act in combination with the autoimmune process responsible for the clinical manifestation of MS. Smoking, alcohol abuse, pneumonia and influenza, urinary tract infections, heart disease and cancer were all predictive factors of reduced survival among this patient group. Further work is warranted to evaluate how these factors compare with those in the general population.