Multiple sclerosis (MS) is a chronic neurological condition that often causes disability and, in some cases, leads to premature death. Mortality data constitute one of the most important resources available to epidemiologists for monitoring the health of populations as well as representing a resource to generate hypotheses about disease aetiology.1 Important information derived from mortality studies in MS to date have included an association with latitude,2 increased risk in women3 and racial differences in disease frequency.3

Nationally and regionally acquired mortality statistics collected through the use of death certificates are the most widely available source of data and are commonly used to study temporal patterns of disease,4^–8 document common causes of terminal comorbidity and to estimate prevalence data for allocation of healthcare resources.9 However, patients with MS may die either from the direct effect of the disease, or unrelated causes; the majority of studies reporting approximately 60–70% of deaths being attributable to the disease itself or its complications.10^–17 As a result, documentation of MS on death certificates is highly variable, ranging between 73% and 94%.1 16 18^–21 The effect of these factors is that epidemiological statistics for MS derived from national coded registers may significantly underestimate mortality, be selective for certain patient subgroups1 and be unrepresentative of the prevalent population with MS.

Although a large number of epidemiological studies relating to MS have been published, there is a relative paucity of data in relation to mortality and cause of death. Some short-term studies have attempted to estimate survival. However, the duration of follow-up and the number of patients dying as a proportion of the population in question are low and therefore frequently inadequate to determine factors related to outcome.13 21 22 Studying death certificates in a prospective cohort of patients known to have MS is a valuable data source for establishing accurate mortality data and also provides important information on the accuracy and utility of death certificates. This study examines a representative population-based cohort of patients with MS followed prospectively for more than 20 years, resulting in a larger number of deaths and hence more precise estimates of survival. The aims of this study were to determine survival from disease onset, examine causes of death, identify whether deaths were related to MS or its complications, assess the frequency of MS being represented on death certificates for MS (and hence utility for prevalence estimates in the UK) and determine the relative mortality of male and female MS patients compared with the general population.

METHODS

A population-based survey was performed in South Glamorgan, Wales, in 1985 and identified patients from a number of sources, including a departmental index, hospital activity analysis, general practices, the MS Society membership, community nurses and physiotherapists.23 Patients were included if they were resident within the county of South Glamorgan, alive on prevalence day and fulfilled diagnostic criteria for MS. This study demonstrated a prevalence of 441/376 718 (117/10⁵) on 1 January 1985, 86% of whom had clinically definite or probable MS according to criteria proposed by Poser.24 Fourteen per cent of patients were classified as having suspected MS, which included patients who would previously have been classified as having MS by earlier diagnostic criteria but did not fulfil revised criteria for definite or probable disease. All cases were flagged with the Office of Population Censuses and Surveys, and copies of death certificates collected prospectively. Death certificates were examined to determine cause of death, age at death and whether MS was the underlying cause of death.

Those patients in whom no death certificate had been received were traced to ensure current vital status. The hospital patient administration system was used to determine the last known address of the patient and their most recent general practitioner (GP) was contacted to confirm residence and status, and/or the National Health Service Administrative Register (NHSAR) was used to trace patients still resident in Wales. The NHSAR is a database that holds demographic data for all individuals registered with a Welsh GP. Data held include NHS number, name, address, date of birth, date of death and GP registration details. The current database records 11 years of demographic information on patients.25

On the death of an individual in England and Wales, a death certificate is completed by a registered medical practitioner who has attended the deceased during their last illness, and the cause of death recorded. If this is unknown, the death is referred to the Coroner who will provide a certificate. The death certificate contains two parts for documentation of the cause of death; Part 1 is further subdivided into sections a, b and c. In Part 1a, the disease or condition leading directly to death is recorded. In Part 1b, other conditions (if any) leading to entries in 1a are recorded, and in Part 1c other conditions or diseases leading to entries in Part 1b (if any) are recorded. Part 2 documents any other conditions that have significantly contributed to death but not directly related to the disease or condition causing it. Death certificates are used to code the cause of death using the International Classification of Diseases (ICD); the 10th version of the ICD has been in use since 1993.

Cause of death was deemed to be due to MS or its complications if MS appeared in Part 1a, 1b or 1c, or if related bronchopneumonia if MS was present in Part 2. These were included as many of the patients appeared to die at a premature age and therefore this was felt to be related to MS. Following individual case note review, a few cases in which death was thought to be clearly related to MS, but MS only appeared in Part 2 of the death certificate or did not appear at all, were also included as MS-related deaths. Examples of these were acute peritonitis due to a leak around a feeding tube in a patient with an EDSS score of 9.5, and acute pulmonary bronchopneumonia or embolism secondary to MS-related immobility (n = 6).

Statistical methods

Differences in proportions or continuous variables were tested using the χ² test or analysis of variance. The cohort was followed up from 1 January 1985 until 27 October 2006 (date of censoring). We determined the mortality experience of our cohort by calculating mortality rates per 1000 person-years, taking into account censoring for patients that were still alive at the end of follow-up. This was undertaken for all-cause mortality and then repeated for cause-specific mortality rates, infectious diseases, cancer, cardiovascular disease and respiratory disease. The sex-specific rate ratios were calculated using Poisson regression models accounting for the person-years at risk. To control for age differences, we grouped age into three categories (15–54, 55–74 and 75+ years).

We also compared the mortality experience of our cohort with what would have been expected from the general population. As the general population mortality rates have changed over the follow-up period, we partitioned the person-years at risk for each patient by 5-year age bands (20–24, 25–29, 30–34, etc up to 90+ years) and by 5-year calendar periods (1985–1989, 1990–1994, etc) stratified by sex. For example, if a female patient was aged 23 years in 1985 and died aged 29 years in 1991, they would contribute 2 years to the 20–24-year age group in the calendar period 1985–1989, 3 years to the 25–29-year age group in the calendar period 1985–1989, and 2 years to the 25–29-year age group in the calendar period 1990–1994. This dataset was then merged with the external mortality rates for England taken from the mid-point of each time period (eg, 1987, 1992, 1997, 2002) by calendar period, gender and age groups. We used the 2002 rates for the last calendar period 2005–2009. The expected number of events expected if patients with MS had the same mortality experience could then be calculated. The standardised mortality ratio (SMR) is the ratio of observed divided by expected deaths, which is usually multiplied by 100 to convert it to a percentage. We have decided to present it as a simple ratio to avoid generating large numbers. Hence, if we observed 20 deaths but would have only expected 10 deaths, our SMR would be 2 (rather than the conventional 200), implying that our patients were twice as likely to die than the general population in relative terms. This method controls for any differences in age and sex structure between our patient cohort and the general population and adjusts for secular trends in mortality risk. We compared the sex-specific rate ratios using Poisson regression models by taking the expected number of events rather than person-years at risk. All analyses were performed using Stata 10.

RESULTS

Of 441 patients identified in the original prevalence study, 379 were initially considered for analysis; 62 were excluded, of whom 46 had suspected disease in 1985 with no further events or re-investigation, and 16 in whom the diagnosis had been revised. In total, 259/379 (68.3%) patients were women and 120/379 (31.7%) men; 148 (39.1%) patients remained alive, 221 (58.3%) had died and 10 (2.6%) were untraceable, so were dropped from further analysis.

Median survival time was 33.0 years (interquartile range 21.1, 42.7 years) from date of diagnosis and 38.0 years (IQR 27.3, 48.7 years) from symptom onset. In total, 147/255 (57.7%) women and 74/114 (64.9%) men had died (p = 0.19). Mean age at death was 65.3 years (95% CI 63.4 to 67.1, range 25.0 to 101.0, SD 14.3). Mean age at death was 65.3 years (SD 14.8) for women and 65.2 years (SD 13.3) for men. “MS” was present in Part 1a of the death certificate in 20 (9.0%), Part 1b or c in 75 (33.9%) and in Part 2 in 61 (27.6%). In 27% of patients, MS was not mentioned in any part of the death certificate.

Cause of death

Cause of death was related to MS in 128 (57.9%) and unrelated in 93 (42.1%) patients. The most common cause of death was respiratory disease, which accounted for 47.5% of all deaths in the study. Cardiovascular disease accounted for 16% of deaths and was more frequent in men than women, as would be expected in the general population (22.5% vs 12.9%). One accidental death and no suicides were recorded (table 1). There was no strong evidence that there were sex-specific differences in the proportion of deaths (p = 0.07). Mean age at death was lower for patients in whom primary cause of death was coded as MS compared with other causes with cardiovascular disease having the highest mean age at death. Overall, there was a significant difference in age at death between different causes of death (p-value for heterogeneity <0.001).

MS-related deaths

There was a significant difference in the proportion of MS-related deaths between men and women: 34 (45.9%) male deaths related to MS compared with 94 (63.9%) female deaths; 18% more deaths were related to the disease in females than males (95% CI 4.2 to 31.1) (p = 0.01) (table 2). Mean age at death in patients who died of MS-related causes was 62.5 (SD 15.0) years compared to 69.3 (SD 12.3) years for unrelated deaths, making those dying of MS or related complications, on average, 6.9 years younger than those dying from unrelated causes (p<0.001). Those dying of MS-related causes had a significantly younger age at disease onset (32.5 years) compared with those dying of unrelated causes (36.8 years) (p = 0.01).

Differences in those with MS present on death certificate compared to absent

There was no significant difference overall in the proportion of males with MS recorded on their death certificate (53, 71.6%) compared with females (104, 70.7%) (p = 0.89) (table 3). However, “MS” appeared on the death certificate in only 33 of 93 (35%) of those who died of non-MS related causes (table 2).

There was a significant difference in the proportion of males having MS recorded on the death certificate for non-MS-related deaths compared with females, with a difference of 16.2% (95% CI 5.8 to 27.8) (p = 0.04) (table 3). In addition, patients with MS recorded on the death certificates died at a younger age (p = 0.03) and also had an earlier disease onset (p<0.001). No difference was identified in mean disease duration for those with and without MS on the death certificate (p = 0.15).

Mortality rates and standardised mortality ratios

Three further patients were dropped from analysis as they were censored prior to 1 January 1985, leaving 366 subjects contributing a total of 218 events (all-cause) and 5609 person-years at risk, with a median follow-up period of 18.5 years (range 0.11 to 21.8 years). The crude mortality rate was 38.9 per 1000 person years (all-cause), with the highest rate being for respiratory deaths (18.6 per 1000 person-years) (table 4). Male patients with MS had slightly greater mortality rates than women (43.2 per 1000 versus 37.1 per 1000) (see fig 1), but this was consistent with chance (p = 0.15) and they were more likely to die of cancer (relative rate 3.14, p = 0.01) and cardiovascular disease (relative rate 2.45, p = 0.01).

The SMRs were elevated for all-cause (2.79), infectious disease (29.6) and respiratory mortality (11.7), whereas the deaths rates from cancer and cardiovascular disease were consistent with expected national rates. Men were relatively less likely to have died from all-causes (28%) and respiratory disease (40%) compared with women, given the differences in sex mortality risks for the general population. In contrast, although more male patients with MS died of cardiovascular disease, this was totally consistent with the sex differences in the general population for cardiovascular disease so that the sex relative risk showed almost no difference.

DISCUSSION

This study has followed a population-based cohort for more than 20 years by prospectively collecting death certificates. It has enabled accurate contemporary estimates of survival and disease duration for patients with MS in the UK, together with a practical examination of the utility of death certificates and documentation of common causes of terminal comorbidity.

Epidemiological studies of MS prior to 1950 reported mean survival of between 12 and 17 years from onset.26^–30 However, many of these earlier studies from which these figures were derived were conducted using hospital-based patient series and were therefore likely to preferentially identify the most severely affected patients. These initial studies were also performed in the pre-antibiotic era and only included analysis of patients who had died. This is likely to underestimate survival, as those dying early are likely to be those with the most aggressive disease. More recent studies utilising survival analysis, and taking into account survival of those that have not yet died,12 15 17 31^–33 have demonstrated that survival appears to have increased over time and is now likely to be between 24.5 and 36 years from disease onset.12 33 34 It is probable that our median survival of 38 years is related to our long follow-up period, as deaths in studies with shorter follow-up will over-represent patients with more aggressive disease. Apart from selection bias, other potential reasons for prolonged survival include increasing longevity of the general population, more effective treatments and improved access to healthcare resources.

Despite the observation of increasing disease duration, we have continued to demonstrate a pattern of premature death in patients with MS, who have a relative mortality risk of almost three times that of the general population. Women with MS do relatively worse, but not because they have more aggressive disease. Although there was almost no sex difference in the cohort mortality rates between male and female patients, the SMR was greater for women than for men. Because women in the general population have longer life expectancy and lower age-specific mortality rates, male patients with MS had almost a 30% relative reduction in SMRs than women. This was also seen for deaths from respiratory disease. This is probably explained by sex differences in smoking behaviour and hence chronic obstructive airways disease (COAD) in the general population, with men having higher disease rates than women, leading to chronic respiratory deaths. In an MS cohort, most respiratory deaths will not be due to COAD but an acute respiratory infectious episode superimposed on end-stage disease.

A large Danish study reported an SMR of 2.9, similar to our findings, with an excess death rate of 13.4 per 100 000/year.34 In a clinic-based Canadian study, the SMR was 2.00 overall, with a high proportion of deaths being due to suicide which, if excluded, reduced the SMR to 1.79.

Mortality statistics are commonly used as a source of data for estimating disease frequency but our study suggests that employment of this methodology alone would have the effect of underestimating MS disease prevalence by at least 27%. Furthermore, if only underlying cause of death (ie, Part 1) had been used to identify patients, then an additional 29% of patients would not have been included. Other authors have reported rates of omission for documentation of MS on the death certificate of between 6% and 27%.1 16 18 19 34 The most common reason for failure to record MS on the death certificate is that the disease played no role in the patients’ death, as Part 2 of the death certificate only lists conditions that contribute to the cause of death and not all medical conditions from which the deceased person suffered at the time of their death. The relatively high proportion of patients with MS missing from the death certificate recorded in this study may also be due to the length of follow-up, with those dying at older ages having more competing causes of death listed in place of MS.

Our study also suggests that patients with MS recorded on the death certificate had a younger age at disease onset and death compared to those without, suggesting that if death certificates alone are used to determine characteristics of the disease, both age at onset and death will be underestimated. However, there was no difference in the sex ratio or disease duration. These demographic differences seen in those with and without MS present on the death certificate have not previously been reported in the UK, although a comparable study from the USA does suggest that available mortality data underestimated age at onset of MS, overestimated the female to male ratio and underestimated age at death and disease duration.1

The most common cause of death in our study was respiratory disease or infection followed by cardiovascular disease, occurring in 47.5% and 15.8%, respectively. Respiratory disease has been shown in another study to be the most common cause of death in people with MS12 but others observers conclude that MS is the most common underlying cause of death.35 36 Some studies have also reported significantly elevated rates for suicide among patients with MS34^–36 compared with control populations, one Canadian study reporting from a clinic-based cohort recording suicide as the cause of death in 28.6% of patients with MS.36 In addition, a large study of a Danish MS registry also found an increased risk of suicide in people with MS with an SMR of 2.12 and 3.15 within the first year of diagnosis.37 However, one other study also failed to identify an excess suicide rate.17 In our study, there were no deaths due to suicide and only one due to accident. There are a number of reasons why this marked differential rate of suicide between studies may exist. First, clinic-based cohorts may include more severely affected individuals who may have higher rates of affective disorders and are not representative of the prevalent MS population. Many studies have shown an association between high levels of disability and depression,38^–40 and one study showed an increased risk of suicidal ideation in patients with higher levels of physical disability.41 Despite this, high levels of suicide have also been recorded in community-based samples of patients with MS.38 Second, in the study from Canada, cause of death was only determined in 82.1% of cases, perhaps leading to an overestimation in the proportion of suicide. Third, there is considerable variation in suicide rates, both geographically and ethnically; for example, rates in the general population of Canada for 2000 are 18/10⁵ for men and 5/10⁵ for women42 compared with 12.8/10⁵ and 2.5/10⁵ in Wales for the same time period.43 However, it is likely that these points are not sufficient to account for all of the difference observed and other, as yet unidentified, factors may also play a role.

In conclusion, we show that mortality is increased in both men and women with MS compared to the general population, although women with MS lose relatively more years of life than men. Our observed median survival of 38 years is longer than previous studies and suggests that past publications may have been biased towards more severe cases, secular improvements in care or that MS is presenting with a milder natural history compared with past series. A significant proportion of patients do die of MS or related complications, but nearly one-third of people dying with MS have no mention of the disorder on their death certificate. In addition, those patients with MS recorded on the death certificate die at a younger age than those without. These results have implications for the use of nationally coded statistics in the analysis of epidemiological trends in MS.