- Email: [email protected]
Factors influencing mortality in rheumatoid arthritis
J Chron Dis Vol. 39, No. 2, pp. 137-145, 1986
0021-9681/86$3.00+ 0.00
Printed in Great Britain. All rights reserved
Copyright G 1986Pergamon PESS
FACTORS INFLUENCING MORTALITY RHEUMATOID ARTHRITIS
Ltd
IN
D. P. M. SYMMONS,** P. PRIOR,’ D. L. SCOTT,’ R. BROWN’and C. F. HAWKINS* ‘Cancer Epidemiology Research Unit, University of Birmingham and 2Rheumatism Research Wing, The Medical School, University of Birmingham, Birmingham, B15 2TJ, U.K.
Abstract-The prognosis in rheumatoid arthritis respecting mortality was studied in a consecutive series of 489 hospital patients over a period of 18 years. The relative risk of mortality was raised in both men (2.6; p < 0.001) and women (3.4;p < 0.001).In the women the relative risk was also influenced by prior duration of RA and was characterised by a diminution in risk 5-9 years after first presentation. Relative risks for men were more uniformly distributed over time. Annual excess mortality rates were strongly associated with age at first presentation in women, the rate increasing with increasing age in both the group seen within 5 years of onset of disease (x:,, for trend = 30.4; p < 0.001) and in the later referral group (& = 34.0; p < 0.001). A similar but much less marked effect was observed in men in the early referral group (Q,, * - 13.7., p < 0.001)only. These results suggest that initially women may have a milder form of disease and that hormonal status may affect prognosis. Future long-term therapeutic studies in RA should take into account the prognostic factors of age, sex and duration of disease.
INTRODUCTION
studies have now established that patients with rheumatoid arthritis have a reduced life expectancy [l-9]. However the possible factors contributing to the excess mortality and their effect on the prognosis of the disease have not been fully explored. We have studied the effect of sex, age and presentation and duration of disease on the outcome of rheumatoid arthritis. The information used for these analyses was gathered during a study on cause of death [9] and cancer morbidity [lo] in a series of 489 patients from Queen Elizabeth Hospital, Birmingham. MORTALITY
Patients The series consisted of 489 consecutive patients (both inpatients and outpatients) with rheumatoid arthritis (probable, definite or classical by ARA criteria) seen by the rheumatology staff of the Queen Elizabeth Hospital (QEH) between 1964 and 1978. All were under the care of one physician (CFH). The patients were identified from a register of attenders begun in 1964. A few patients had first attended QEH before 1964. They represent survivors from an unknown cohort and have been entered into the analysis at 1 January 1964. All other patients were entered into the analysis at the date of first attendance at the index hospital (FAIH). Patients were followed until the date of death or 3 1st December 1981 through clinical notes, the Birmingham Regional Cancer Registry and the National Health Service Central Register. Copies of all death certificates were obtained. Information on 41 cases (8.4%) was insufficient to confirm the diagnosis or for tracing. Of the 448 entered into the analysis the final status of 433 (96.6%) was verified. *Author for correspondence. 137
D. P. M.
138
SYMMONS et al.
At the time of the study there was a shortage of rheumatologists in the Birmingham area. The patients referred to QEH came from a wide area but represent only a small proportion of the total rheumatoid population. The main reasons for referral were: (a) for definitive diagnosis and help with management disease
early in the course of the
(b) for management of refractory established disease.
or associated conditions in
and RA, its complications
Disease severity varied widely. Many patients had mild disease and were only seen on one or two occasions. In an attempt to dissociate those seen early in the disease from those seen later because of complications the patients were divided into 2 groups: Group 1: those first seen within 5 years of the onset of RA; Group 2: those first seen more than 5 years after onset. The distribution of the series by sex, age at first attendance and duration of disease is shown in Table 1. METHODS
The data were examined by three methods to describe and evaluate the pattern of mortality in a defined series of patients. 1. Relative risk (or standardised mortality ratio) Age- and sex-specific mortality rates were computed from mean annual deaths (1969-73) [l l] and census population figures for 1971 [12] for England and Wales. Patients were traced to 31st December 1981. The resultant person-years at risk were multiplied by the relevant specific mortality rates to compute the number of deaths that might be expected to occur during the period of review. The expected numbers of deaths were computed at each year of follow-up by sex and age and by duration of disease at presentation. The significance of the differences between observed and expected numbers were tested by use of the Poisson distribution, the level of risk being expressed in relative terms: relative risk (RR) = observed/expected
number of deaths
Differences between pairs of subgroups were assessed by a X2-statistic with Yate’s correction. Interval- and age-groups were assessed by X2-statistics for heterogeneity and trend. 2. Life table survival curve Conventional life table methods for censored data were used to obtain the observed cumulative probability of survival (po). The cumulative expected probability of survival &) was derived from the same censored person-years of exposure and the number of TABLEI.
RA: ACE D,STR~UTION OF THE SERES Number
Age at first attendance
Group 1 onset < 5 years Males Females
of patients Group 2 onset > 5 years Males Females
<20 20-29 30-39 40-49 5&59 60+
5 I3 I9 29 22
II I9 26 43 32 26
2 4 IO I4 I9 iI
2 8 II 35 45 35
Total
95
157
60
136
51
45
50
56
Median
age (Yr)
7
Factors
Influencing Mortality in RA
139
TABLE 2. RA: MORTALITY BY REFERRAL GROUP Patients Group
Sex
I
0
E
O/E
P
EMR
19.78 16.71 36.50
2.4 2.6 2.5
*** *‘a
2.6 1.4 I.8
(N = 253)
T F
47 44 91
Group 2 (N = 196)
M F T
35 73 108
12.09 18.15 30.24
2.9 4.0 3.6
*** l ** *‘*
4.0 3.7 3.8
TOtLdl
M F T
82 II7 199
31.87 34.86 66.73
2.6 3.4 3.0
l
** *** ***
3.1 2.4 2.6
(N = 449)
M
a’*
0: Observed number of deaths. E: Expected number of deaths. EMR: Excess mortality (%/yr). ‘p < 0.05; **p< 0.01; ***p< 0.001.
deaths expected to occur in each year of follow-up. Relative survival curves (po/pE), which are, therefore, corrected for the varying age-distributions of survivors, have been plotted with 95% confidence limits [13, 141to illustrate the cumulative net survival experienced by the series. Thus 100 -pO/pE represents the percent excess mortality that might be attributable to the presence of RA. 3. Excess mortality rate (EA4R) Excess mortality rates for specific groupings are also presented. EMR = observed - expected number of deaths x lOO/“Person years” at risk Differences between pairs of subgroups were assessed by means of a t-test based on the binomial distribution. Trends for rates were evaluated from equations suggested by Rothman and Boice [15].
RESULTS
The results of the analyses are summarised by referral group and sex in Table 2. Group 1 In comparison with rates in the general population the overall relative risk of mortality was highly significant (RR = 2.5; p < 0.001; 95% confidence limits (CL) = 2.0-3.1). The relative risk in women was marginally higher than in men but not significantly so (of,, = 2.82). The annual excess mortality rate (EMR) in men was, however, marginally higher than in women (JJ < 0.05), the difference being 1.2%/yr (95% CL = 0.20-2.3). The mean EMR for the group was 1.8%/yr (95% CL = 1.2-2.6). Group 2
Patients first seen more than 5 years after onset of RA also experienced a higher relative risk of mortality (RR = 3.6; p < 0.001; 95% CL = 3.0-4.4). Again, the difference between relative risks in men and women did not achieve the 5% significance level (x:,, = 2.28). In Group 2 patients the annual excess mortality rates were similar for men and women (EMR = 3.8%/yr; 95% CL = 2.94.9). Group 1 us Group 2
Overall the relative risk was higher in Group 2 (&, = 6.46; p -C0.05) but the effect was due mainly to the increased relative risk in women (xt,, = 4.59; p < 0.05) and was only marginally reduced by stratification for age (x t,, = 4.23; p c 0.05). This result suggests that the effect from the duration of RA before presentation is largely independent of age. The difference in relative risks for men was not significant (xt,, = 0.62).
D. P. M.
140
Group I
Group 2
‘p < 0.05;
SYMMONS et al.
O-4
M F T
17 16 33
5.93 4.57 10.50
2.9 3.5 3.1
‘** *** ***
2.7 1.6 2.0
5-9
M F T
II 7 I8
6.43 5.58 12.10
I.7 1.3 1.5
-
1.3 0.2 0.6
lo+
M F T
I9 21 40
7.43 6.56 13.99
2.6 3.2 2.9
*** *** ***
4.0 2.6 3.1
c&4
M F T
13 31 44
3.90 5.93 9.83
3.3 5.2 4.5
*** *** ***
3.5 4.2 4.0
5-9
M F T
I3 IO 23
4.10 6.21 IO.31
3.2 I.6 2.2
*** ,***
5.0 0.8 2.0
lo+
M F T
9 32 41
4.09 6.00 10.09
2.2 5.3 4.1
* *** ***
3.9 6.3 5.7
**p < 0.01; l**p < 0.001.
The effects of duration of disease after presentation are given in Table 3 for three time-periods by sex. In both referral groups the women showed a lower relative risk in the middle period (5-9 years). After stratifyng for age the differences between periods remained significant (& = 6.46; p < 0.05 and x f2)= 7.1; p < 0.05 for Groups 1 and 2 respectively). The effect was not found in men (x1*) 2 - 1.85 and 1.05 for Groups 1 and 2 respectively). In women the EMR’s showed an even more pronounced fall during 5-9 years. The significance of the difference in the distribution of rates over time was remarkably high even after stratification for age (x T2,= 17.73; p < 0.001 and 29.46; p < 0.001 for Groups 1 and 2 respectively). Although the results for men Group 1 showed a similar pattern, no significant difference could be demonstrated (xt2, = 5.57). In Group 2 the rates for men were high in all three periods. The effect of age at presentation on the risk of mortality was diverse (Table 4). Looking first at relative risks: although women first seen under the age of 40 years had apparently low relative risks, no heterogeneity for age was found for either men or women in Group 1 (x&, = 0.94 and 1.08 for men and women respectively after stratification for interval). In Group 2 the relative risk decreased with increasing age in men (xi,, for trend = 4.41; p < 0.05) but not in women, When excess mortality rates are considered, however, a remarkably strong effect of age at presentation on mortality was found. EMR’s increased with increasing age in women in both Group 1 and Group 2 (x$, for trend = 30.39 and 34.01 respectively; p < 0.001). Stratification for interval marginally increased the association. A similar, but smaller, effect was found in men in Group 1 but not in Group 2 (xi,, for trend = 13.66; p < 0.001 and 0.36 respectively). Life tables
Although the net cumulative percent survival in women in Group 1 appears higher than that for men, there is substantial overlap of the 95% confidence limits [Fig. 11. However, a test between the mean annual excess mortality rates-2.6% for men and 1.4% for women-showed the difference to be of marginal significance (t = 2.35; p < 0.05). The less regular curves obtained for Group 2 patients [Fig. l] reflect the greater heterogeneity of a late referral series, and the fluctuation of the curves about each other suggest that the mortality attributable to the presence of the disease is similar for men and women. Group 2 started with fewer entrants and experienced a higher rate of mortality
Factors
TABLE4. RA:
Influencing
MORTALS
Age at FIAH (yr)
Patients
BY
I
AGE
141
FAIH AND REFERRAL GROUP
0
E
OIE
M
0.75 1.11 1.86 10.43 8.99 19.42
4.0 2.7 3.2 2.4 2.2 2.3
‘* l ** 1.1 ***
0.6 0.3 0.4 2.1 I.1 1.7
P
EMR
4&59
M F T
3 3 6 25 20 45
60+
M F T
I9 21 40
8.61 6.62 15.23
2.2 3.2 2.6
*** *‘* l **
1.4 6.7 7.0
M F T
5
c40
6
0.46 0.56 I .02
10.9 1.8 5.9
** ***
2.6 0.1 1.0
F T
20 39 59
5.14 8.92 14.06
3.9 4.4 4.2
*** *** ‘**
5.4 3.2 3.9
M F T
IO 33 43
6.49 8.69 15.18
I.5 3.8 2.8
*** l **
3.2 10.3 8.0
F T
M
Group 2
AT
in RA
Sex
c40
Group
Mortality
4&59
60+
I
l
l
‘p < 0.05; **p < 0.01; p < 0.001
leading to only small numbers of survivors in the later years and, consequently, wider confidence limits at the end point. The mean annual excess mortality rates were similar-2.8% for men and 3.0% for women. In both Groups the curves for women are characterised by a plateau-effect between 5 and 9 years of observation which correspond with the lower relative risks and EMR’s noted in Table 3. Prognosis in young patients
Particular attention was paid to deaths occurring in the 120 patients first seen under the age of 40. Overall the relative risk of mortality is high in this group (RR = 4.2; p < 0.001; 95% Cl = 2.2-7.3) but no difference could be detected between men and women. There
Intmal
100
2
90
;
00
2z
70
z
60
(Yeers
from FAlH )
)
100
F1c.1. Cumulative Group 2 patients
c;
90
;
80
g
70
z u)
60
E z ;= =
5o 40 30
relative survival rates with 95% confidence limits for Group (bottom), showing results for men (-) and women (------)
I (top) and separately.
142
D. P. M.
SYMMONS
et al.
was, however, a significant difference between EMR’s for men and women (t = 2.66; p < 0.01). This effect was due mainly to the higher rate in men in the late referral group (EMR = 2.6) compared with that in women (EMR = 0.14) (Table 4). Although the numbers are small, with an excess of 9 deaths out of a total of 12 observed which might be attributable to RA, the clinical cause of death and treatment received by the 12 patients in this group were considered to be of special interest and are summarised in Table 5. Ten of these 12 patients (83%) had received steroids compared with 50% of the group as a whole (x :,, = 4 .08., p < 0.05). The high proportion receiving steroids and second-line drugs partly reflects the fact that this group of young patients had particularly severe diseases but it remains possible that therapy contributed to some of these deaths. RA was given as the underlying cause of death on only 3 death certificates. In retrospect the first patient was considered to have had Still’s disease. Two out of the three patients dying with amyloidosis in the whole series were in this group of young patients.
DISCUSSION
Rheumatoid arthritis is a chronic disease frequently causing progressive disability over several decades. Many patients with RA will never be referred to hospital and those who do attend may be referred for a variety of reasons. This inevitably leads to bias in the selection of patients seen in hospital. Some patients in our series were referred for confirmation of the diagnosis, but most attended for uncontrolled active RA or for complications occurring at varying times after diagnosis. This staggered timing of presentation raises difficulties in assessing the prognosis of the disease as a whole. Taking an arbitrary cut-off time of 5 years we have examined separately those patients seen early in their disease (within 5 years of onset) and those seen later. It was felt that those patients attending later were more likely to have developed complications leading to their referral to hospital. This procedure is validated by the finding that Group 2 patients experienced a higher relative risk of dying than Group 1 patients and the main differences were due to the complications of infection and urinary diseases [9]. In order to evaluate the impact of a disease on mortality a basis for comparison is required. Some studies have used case-control methods [7,8]; many have shown only crude mortality rates and others have presented survival curves with [l, 5, 61 and without [2, 31 a comparative curve from the general population. The implications of these differing methods have been discussed previously [16]. In this study we have compared mortality, in men and women separately, with that in a general population of similar age. Thus the relative risk of dying for women was found to be higher than for males when compared with persons of their own sex. Relative risk indicates the proportion of the observed deaths that might be attributed to the disease and to possible confounding factors. Life tables and EMR analyses are perhaps more useful in the clinical context since they allow comparisons between the sexes and other sub-groups, again in relative terms-i.e. corrected for “expected” deaths. These methods express mortality in terms of the proportion of patients in whom death might be attributed to the presence of RA. Mortality rates for England and Wales were used to compute the expected number of deaths whereas patients were drawn from the West Midlands Region. In genera1 the Region is reasonably representative of the country as a whole covering both urban and rural areas. Standardised Mortality Ratios (SMR) for the Region were 104 for males and 102 for females (1969-73) and were 5th in rank out of 9 Regions. The result of correcting for this possible bias would result in increasing the overall expected number by approximately 2 deaths (0.97 for men and 1.08 for women). Such a correction would not have an appreciable effect on the given significance levels. Socio-economic factors might also introduce bias in analyses of mortality but those effects are difficult to investigate in retrospective studies. In broad terms, the class
C
16 13 17 18 33
45
47 50 53 57
M
M M M F
5 6 7
8
9 IO II 12
S: Steroids. G: Gold. C: Chloroquine or hydroxychloroquine P: o-Penicillamine. D: Dapsone. X: Radiotherapy. WC: Inferior vena cava. GI: Gastrointestinal.
G X
4 6 19
34 38 42
F M M
3 4
D G G
G
-
G
20 14
29 34
M F
2
P
C
Cause of death Acute myocarditis Pulmonary embolus from IVC thrombosis GI haemorrhage Renal failure due to amyloidosis Acute my&id leukaemia Suicide Cerebral Hamartoma Hypertension Perforated diverticulum Septicaemia Myocardial infarction Gastric ulcer Lobar pneumonia Staphylococcal septicaemia Septic arthritis
G G
C
Treatment
9 12
23 28
F M
I
S
Duration of disease Onset-death (yr)
Age at death
Sex
Patient
Multiple sclerosis Amyloidosis
Other conditions
5’
F
D. P. M. SYMMONS et al.
144
distribution in men was not inconsistent with that in the general population. In women there was a trend towards upper social class but it was not clear whether this was due to real effect or to an “up-grading” effect on the reporting of occupation on some death certificates [ 171. The medical record data were not sufficient to allow for a comparison of the groups in terms of therapy or smoking habits. Several aspects of the results suggest that there was in fact a real difference between men and women in their response to RA. First, although the relative risk of mortality was high for both sexes women showed more variation between referral groups and between intervals after presentation. Similar effects were found for excess mortality rates in women and, for both measures (RR and EMR), the effects were independent of age. These results suggest that initially women may have milder disease but with time this advantage is lost. The excess risks in men are distributed more uniformly over time. Second, the proportion of women dying from causes attributable to RA or its complications is strongly associated with age at presentation, the trends of increasing mortality with increasing age, being highly significant for both referral groups. The highest EMR (lO%/yr) is found in Group 2 women who were first seen after the age of 60 in whom the two effects-of age and duration-might be expected to have the greatest impact. Although the effect of age was also found in men in Group 1 the trend was much less marked. Thus, although RA may be more common in women, there may be a protective effect in younger women against its more severe complications. Thus, benefit may be hormonally mediated as it seems to be lost after the menopause. Recent reports suggest that users of oral contraceptives [18,19] may be protected from developing RA. Thus, there appears to be evidence for hormonal status not only influencing who may develop RA but also its ultimate prognosis. Although it might be considered that differences between men and women demonstrated here are due to the disparity of numbers (male:female ratio = 1: 1.8), the fact that the numbers of deaths observed and expected are comparable supports the conclusions of real differences between the sexes. These differences, particularly the more pronounced variation in risk in women, have implications for the validity of future long-term therapeutic studies in RA.
Acknowledgements-We thank the Endowment Fund of the Central Birmingham Health Authority, the Cancer Research Campaign and the Arthritis and Rheumatism Council for their financial support. We also acknowledge the help of staff at the National Health service Central Register for long-term tracing of patients.
REFERENCES I.
Cobb S, Anderson F, Bauer W: Length of life and cause of death in rheumatoid arthritis. N Engl J Med 249: 553-556, 1953 2. Duthie JJR, Brown PE, Truelove LH, Baragar FD, Lawrie AJ: Course and prognosis in rheumatoid arthritis: A further report. Ann Rheum Dis 23: 193-202, 1964 3. Uddin J, Kraus AS, Kelly HG: Survivorship and death in rheumatoid arthritis. Arth Rheum 13: 125-130, 1970 4. 5. 6. 7.
Monson RR, Hall AP: Mortality among arthritics. J Chron Dis 29: 459467, 1976 Allebeck P, Ahlbom A, Allander E: Increased mortality among persons with rheumatoid arthritis, but where RA does not appear on the death certificate. Stand J Rheum 10: 301-306, 1981 Allebeck P: Increased mortality in rheumatoid arthritis. Stand J Rheum 11: 81-86, 1982 Isomlki HA, Mutru 0, Koota K: Death rate and causes of death in patients with rheumatoid arthritis. &and J Rheum 4: 205-208,
8. 9.
1975
Koota K, Isomaki H, Mutru 0: Death rate and causes of death in rheumatoid arthritis patients during a period of five years. Sand J Rheum 6: 241-244, 1977 Prior P, Symmons DPM, Scott DL, Brown R, Hawkins CF: Cause of death in Rheumatoid Arthritis. Br J Rheum 23: 92-99,
1984
10. Prior P, Symmons DPM, Hawkins CF, Scott DL, Brown R: Cancer morbidity in Rheumatoid Arthritis. Ann Rheum Dis 43: 128-131, 1984 Office of Population Censuses and Surveys: The Registrar General’s Statistical Reviews for the years 1969-1973. London: HMSO 12. Office of Population Censuses and Surveys: The Registrar General’s Revised Estimates of the Population of England and Wales, 1961-71. London: HMSO, 1975 13. Ederer P, Axtell LM, Cutler SJ: National Cancer Institute (USA) Monograph No. 6, 1961 II.
Factors Influencing Mortality in RA
145
14. Peto R, Pike MC, Armitage NE ef al: Design and analysis of randomised clinical trials. II Analysis and examples. Br J Cancer 35: l-39, 1977 15. Rothman KJ, Boice, Jr JD: Epidemiological Analysis with a Programmable Calculator. Washington, USA: National Institutes of Health, 1979 16. Prior P: Cancer and rheumatoid arthritis: Epidemiological considerations. Am J Med 78(lA): 15-21, 1985 17. Prior P, Symmons DPM: Rheumatoid arthritis and social class. Br J Rheumatol 23: 309-310. 1984 18. Lingrave S, Kay CR. Reduction in incidence of rheumatoid arthritis associated with oral contraception. 19.
Lancet i: 569-57 I, 1978 Vandenbrouke JP, Valkenburg HA, Boerma JW et al. Oral contraceptives and rheumatoid arthritis. Further evidence for a preventive effect. Lancet ii: 839-842, 1982
0021-9681/86$3.00+ 0.00
Printed in Great Britain. All rights reserved
Copyright G 1986Pergamon PESS
FACTORS INFLUENCING MORTALITY RHEUMATOID ARTHRITIS
Ltd
IN
D. P. M. SYMMONS,** P. PRIOR,’ D. L. SCOTT,’ R. BROWN’and C. F. HAWKINS* ‘Cancer Epidemiology Research Unit, University of Birmingham and 2Rheumatism Research Wing, The Medical School, University of Birmingham, Birmingham, B15 2TJ, U.K.
Abstract-The prognosis in rheumatoid arthritis respecting mortality was studied in a consecutive series of 489 hospital patients over a period of 18 years. The relative risk of mortality was raised in both men (2.6; p < 0.001) and women (3.4;p < 0.001).In the women the relative risk was also influenced by prior duration of RA and was characterised by a diminution in risk 5-9 years after first presentation. Relative risks for men were more uniformly distributed over time. Annual excess mortality rates were strongly associated with age at first presentation in women, the rate increasing with increasing age in both the group seen within 5 years of onset of disease (x:,, for trend = 30.4; p < 0.001) and in the later referral group (& = 34.0; p < 0.001). A similar but much less marked effect was observed in men in the early referral group (Q,, * - 13.7., p < 0.001)only. These results suggest that initially women may have a milder form of disease and that hormonal status may affect prognosis. Future long-term therapeutic studies in RA should take into account the prognostic factors of age, sex and duration of disease.
INTRODUCTION
studies have now established that patients with rheumatoid arthritis have a reduced life expectancy [l-9]. However the possible factors contributing to the excess mortality and their effect on the prognosis of the disease have not been fully explored. We have studied the effect of sex, age and presentation and duration of disease on the outcome of rheumatoid arthritis. The information used for these analyses was gathered during a study on cause of death [9] and cancer morbidity [lo] in a series of 489 patients from Queen Elizabeth Hospital, Birmingham. MORTALITY
Patients The series consisted of 489 consecutive patients (both inpatients and outpatients) with rheumatoid arthritis (probable, definite or classical by ARA criteria) seen by the rheumatology staff of the Queen Elizabeth Hospital (QEH) between 1964 and 1978. All were under the care of one physician (CFH). The patients were identified from a register of attenders begun in 1964. A few patients had first attended QEH before 1964. They represent survivors from an unknown cohort and have been entered into the analysis at 1 January 1964. All other patients were entered into the analysis at the date of first attendance at the index hospital (FAIH). Patients were followed until the date of death or 3 1st December 1981 through clinical notes, the Birmingham Regional Cancer Registry and the National Health Service Central Register. Copies of all death certificates were obtained. Information on 41 cases (8.4%) was insufficient to confirm the diagnosis or for tracing. Of the 448 entered into the analysis the final status of 433 (96.6%) was verified. *Author for correspondence. 137
D. P. M.
138
SYMMONS et al.
At the time of the study there was a shortage of rheumatologists in the Birmingham area. The patients referred to QEH came from a wide area but represent only a small proportion of the total rheumatoid population. The main reasons for referral were: (a) for definitive diagnosis and help with management disease
early in the course of the
(b) for management of refractory established disease.
or associated conditions in
and RA, its complications
Disease severity varied widely. Many patients had mild disease and were only seen on one or two occasions. In an attempt to dissociate those seen early in the disease from those seen later because of complications the patients were divided into 2 groups: Group 1: those first seen within 5 years of the onset of RA; Group 2: those first seen more than 5 years after onset. The distribution of the series by sex, age at first attendance and duration of disease is shown in Table 1. METHODS
The data were examined by three methods to describe and evaluate the pattern of mortality in a defined series of patients. 1. Relative risk (or standardised mortality ratio) Age- and sex-specific mortality rates were computed from mean annual deaths (1969-73) [l l] and census population figures for 1971 [12] for England and Wales. Patients were traced to 31st December 1981. The resultant person-years at risk were multiplied by the relevant specific mortality rates to compute the number of deaths that might be expected to occur during the period of review. The expected numbers of deaths were computed at each year of follow-up by sex and age and by duration of disease at presentation. The significance of the differences between observed and expected numbers were tested by use of the Poisson distribution, the level of risk being expressed in relative terms: relative risk (RR) = observed/expected
number of deaths
Differences between pairs of subgroups were assessed by a X2-statistic with Yate’s correction. Interval- and age-groups were assessed by X2-statistics for heterogeneity and trend. 2. Life table survival curve Conventional life table methods for censored data were used to obtain the observed cumulative probability of survival (po). The cumulative expected probability of survival &) was derived from the same censored person-years of exposure and the number of TABLEI.
RA: ACE D,STR~UTION OF THE SERES Number
Age at first attendance
Group 1 onset < 5 years Males Females
of patients Group 2 onset > 5 years Males Females
<20 20-29 30-39 40-49 5&59 60+
5 I3 I9 29 22
II I9 26 43 32 26
2 4 IO I4 I9 iI
2 8 II 35 45 35
Total
95
157
60
136
51
45
50
56
Median
age (Yr)
7
Factors
Influencing Mortality in RA
139
TABLE 2. RA: MORTALITY BY REFERRAL GROUP Patients Group
Sex
I
0
E
O/E
P
EMR
19.78 16.71 36.50
2.4 2.6 2.5
*** *‘a
2.6 1.4 I.8
(N = 253)
T F
47 44 91
Group 2 (N = 196)
M F T
35 73 108
12.09 18.15 30.24
2.9 4.0 3.6
*** l ** *‘*
4.0 3.7 3.8
TOtLdl
M F T
82 II7 199
31.87 34.86 66.73
2.6 3.4 3.0
l
** *** ***
3.1 2.4 2.6
(N = 449)
M
a’*
0: Observed number of deaths. E: Expected number of deaths. EMR: Excess mortality (%/yr). ‘p < 0.05; **p< 0.01; ***p< 0.001.
deaths expected to occur in each year of follow-up. Relative survival curves (po/pE), which are, therefore, corrected for the varying age-distributions of survivors, have been plotted with 95% confidence limits [13, 141to illustrate the cumulative net survival experienced by the series. Thus 100 -pO/pE represents the percent excess mortality that might be attributable to the presence of RA. 3. Excess mortality rate (EA4R) Excess mortality rates for specific groupings are also presented. EMR = observed - expected number of deaths x lOO/“Person years” at risk Differences between pairs of subgroups were assessed by means of a t-test based on the binomial distribution. Trends for rates were evaluated from equations suggested by Rothman and Boice [15].
RESULTS
The results of the analyses are summarised by referral group and sex in Table 2. Group 1 In comparison with rates in the general population the overall relative risk of mortality was highly significant (RR = 2.5; p < 0.001; 95% confidence limits (CL) = 2.0-3.1). The relative risk in women was marginally higher than in men but not significantly so (of,, = 2.82). The annual excess mortality rate (EMR) in men was, however, marginally higher than in women (JJ < 0.05), the difference being 1.2%/yr (95% CL = 0.20-2.3). The mean EMR for the group was 1.8%/yr (95% CL = 1.2-2.6). Group 2
Patients first seen more than 5 years after onset of RA also experienced a higher relative risk of mortality (RR = 3.6; p < 0.001; 95% CL = 3.0-4.4). Again, the difference between relative risks in men and women did not achieve the 5% significance level (x:,, = 2.28). In Group 2 patients the annual excess mortality rates were similar for men and women (EMR = 3.8%/yr; 95% CL = 2.94.9). Group 1 us Group 2
Overall the relative risk was higher in Group 2 (&, = 6.46; p -C0.05) but the effect was due mainly to the increased relative risk in women (xt,, = 4.59; p < 0.05) and was only marginally reduced by stratification for age (x t,, = 4.23; p c 0.05). This result suggests that the effect from the duration of RA before presentation is largely independent of age. The difference in relative risks for men was not significant (xt,, = 0.62).
D. P. M.
140
Group I
Group 2
‘p < 0.05;
SYMMONS et al.
O-4
M F T
17 16 33
5.93 4.57 10.50
2.9 3.5 3.1
‘** *** ***
2.7 1.6 2.0
5-9
M F T
II 7 I8
6.43 5.58 12.10
I.7 1.3 1.5
-
1.3 0.2 0.6
lo+
M F T
I9 21 40
7.43 6.56 13.99
2.6 3.2 2.9
*** *** ***
4.0 2.6 3.1
c&4
M F T
13 31 44
3.90 5.93 9.83
3.3 5.2 4.5
*** *** ***
3.5 4.2 4.0
5-9
M F T
I3 IO 23
4.10 6.21 IO.31
3.2 I.6 2.2
*** ,***
5.0 0.8 2.0
lo+
M F T
9 32 41
4.09 6.00 10.09
2.2 5.3 4.1
* *** ***
3.9 6.3 5.7
**p < 0.01; l**p < 0.001.
The effects of duration of disease after presentation are given in Table 3 for three time-periods by sex. In both referral groups the women showed a lower relative risk in the middle period (5-9 years). After stratifyng for age the differences between periods remained significant (& = 6.46; p < 0.05 and x f2)= 7.1; p < 0.05 for Groups 1 and 2 respectively). The effect was not found in men (x1*) 2 - 1.85 and 1.05 for Groups 1 and 2 respectively). In women the EMR’s showed an even more pronounced fall during 5-9 years. The significance of the difference in the distribution of rates over time was remarkably high even after stratification for age (x T2,= 17.73; p < 0.001 and 29.46; p < 0.001 for Groups 1 and 2 respectively). Although the results for men Group 1 showed a similar pattern, no significant difference could be demonstrated (xt2, = 5.57). In Group 2 the rates for men were high in all three periods. The effect of age at presentation on the risk of mortality was diverse (Table 4). Looking first at relative risks: although women first seen under the age of 40 years had apparently low relative risks, no heterogeneity for age was found for either men or women in Group 1 (x&, = 0.94 and 1.08 for men and women respectively after stratification for interval). In Group 2 the relative risk decreased with increasing age in men (xi,, for trend = 4.41; p < 0.05) but not in women, When excess mortality rates are considered, however, a remarkably strong effect of age at presentation on mortality was found. EMR’s increased with increasing age in women in both Group 1 and Group 2 (x$, for trend = 30.39 and 34.01 respectively; p < 0.001). Stratification for interval marginally increased the association. A similar, but smaller, effect was found in men in Group 1 but not in Group 2 (xi,, for trend = 13.66; p < 0.001 and 0.36 respectively). Life tables
Although the net cumulative percent survival in women in Group 1 appears higher than that for men, there is substantial overlap of the 95% confidence limits [Fig. 11. However, a test between the mean annual excess mortality rates-2.6% for men and 1.4% for women-showed the difference to be of marginal significance (t = 2.35; p < 0.05). The less regular curves obtained for Group 2 patients [Fig. l] reflect the greater heterogeneity of a late referral series, and the fluctuation of the curves about each other suggest that the mortality attributable to the presence of the disease is similar for men and women. Group 2 started with fewer entrants and experienced a higher rate of mortality
Factors
TABLE4. RA:
Influencing
MORTALS
Age at FIAH (yr)
Patients
BY
I
AGE
141
FAIH AND REFERRAL GROUP
0
E
OIE
M
0.75 1.11 1.86 10.43 8.99 19.42
4.0 2.7 3.2 2.4 2.2 2.3
‘* l ** 1.1 ***
0.6 0.3 0.4 2.1 I.1 1.7
P
EMR
4&59
M F T
3 3 6 25 20 45
60+
M F T
I9 21 40
8.61 6.62 15.23
2.2 3.2 2.6
*** *‘* l **
1.4 6.7 7.0
M F T
5
c40
6
0.46 0.56 I .02
10.9 1.8 5.9
** ***
2.6 0.1 1.0
F T
20 39 59
5.14 8.92 14.06
3.9 4.4 4.2
*** *** ‘**
5.4 3.2 3.9
M F T
IO 33 43
6.49 8.69 15.18
I.5 3.8 2.8
*** l **
3.2 10.3 8.0
F T
M
Group 2
AT
in RA
Sex
c40
Group
Mortality
4&59
60+
I
l
l
‘p < 0.05; **p < 0.01; p < 0.001
leading to only small numbers of survivors in the later years and, consequently, wider confidence limits at the end point. The mean annual excess mortality rates were similar-2.8% for men and 3.0% for women. In both Groups the curves for women are characterised by a plateau-effect between 5 and 9 years of observation which correspond with the lower relative risks and EMR’s noted in Table 3. Prognosis in young patients
Particular attention was paid to deaths occurring in the 120 patients first seen under the age of 40. Overall the relative risk of mortality is high in this group (RR = 4.2; p < 0.001; 95% Cl = 2.2-7.3) but no difference could be detected between men and women. There
Intmal
100
2
90
;
00
2z
70
z
60
(Yeers
from FAlH )
)
100
F1c.1. Cumulative Group 2 patients
c;
90
;
80
g
70
z u)
60
E z ;= =
5o 40 30
relative survival rates with 95% confidence limits for Group (bottom), showing results for men (-) and women (------)
I (top) and separately.
142
D. P. M.
SYMMONS
et al.
was, however, a significant difference between EMR’s for men and women (t = 2.66; p < 0.01). This effect was due mainly to the higher rate in men in the late referral group (EMR = 2.6) compared with that in women (EMR = 0.14) (Table 4). Although the numbers are small, with an excess of 9 deaths out of a total of 12 observed which might be attributable to RA, the clinical cause of death and treatment received by the 12 patients in this group were considered to be of special interest and are summarised in Table 5. Ten of these 12 patients (83%) had received steroids compared with 50% of the group as a whole (x :,, = 4 .08., p < 0.05). The high proportion receiving steroids and second-line drugs partly reflects the fact that this group of young patients had particularly severe diseases but it remains possible that therapy contributed to some of these deaths. RA was given as the underlying cause of death on only 3 death certificates. In retrospect the first patient was considered to have had Still’s disease. Two out of the three patients dying with amyloidosis in the whole series were in this group of young patients.
DISCUSSION
Rheumatoid arthritis is a chronic disease frequently causing progressive disability over several decades. Many patients with RA will never be referred to hospital and those who do attend may be referred for a variety of reasons. This inevitably leads to bias in the selection of patients seen in hospital. Some patients in our series were referred for confirmation of the diagnosis, but most attended for uncontrolled active RA or for complications occurring at varying times after diagnosis. This staggered timing of presentation raises difficulties in assessing the prognosis of the disease as a whole. Taking an arbitrary cut-off time of 5 years we have examined separately those patients seen early in their disease (within 5 years of onset) and those seen later. It was felt that those patients attending later were more likely to have developed complications leading to their referral to hospital. This procedure is validated by the finding that Group 2 patients experienced a higher relative risk of dying than Group 1 patients and the main differences were due to the complications of infection and urinary diseases [9]. In order to evaluate the impact of a disease on mortality a basis for comparison is required. Some studies have used case-control methods [7,8]; many have shown only crude mortality rates and others have presented survival curves with [l, 5, 61 and without [2, 31 a comparative curve from the general population. The implications of these differing methods have been discussed previously [16]. In this study we have compared mortality, in men and women separately, with that in a general population of similar age. Thus the relative risk of dying for women was found to be higher than for males when compared with persons of their own sex. Relative risk indicates the proportion of the observed deaths that might be attributed to the disease and to possible confounding factors. Life tables and EMR analyses are perhaps more useful in the clinical context since they allow comparisons between the sexes and other sub-groups, again in relative terms-i.e. corrected for “expected” deaths. These methods express mortality in terms of the proportion of patients in whom death might be attributed to the presence of RA. Mortality rates for England and Wales were used to compute the expected number of deaths whereas patients were drawn from the West Midlands Region. In genera1 the Region is reasonably representative of the country as a whole covering both urban and rural areas. Standardised Mortality Ratios (SMR) for the Region were 104 for males and 102 for females (1969-73) and were 5th in rank out of 9 Regions. The result of correcting for this possible bias would result in increasing the overall expected number by approximately 2 deaths (0.97 for men and 1.08 for women). Such a correction would not have an appreciable effect on the given significance levels. Socio-economic factors might also introduce bias in analyses of mortality but those effects are difficult to investigate in retrospective studies. In broad terms, the class
C
16 13 17 18 33
45
47 50 53 57
M
M M M F
5 6 7
8
9 IO II 12
S: Steroids. G: Gold. C: Chloroquine or hydroxychloroquine P: o-Penicillamine. D: Dapsone. X: Radiotherapy. WC: Inferior vena cava. GI: Gastrointestinal.
G X
4 6 19
34 38 42
F M M
3 4
D G G
G
-
G
20 14
29 34
M F
2
P
C
Cause of death Acute myocarditis Pulmonary embolus from IVC thrombosis GI haemorrhage Renal failure due to amyloidosis Acute my&id leukaemia Suicide Cerebral Hamartoma Hypertension Perforated diverticulum Septicaemia Myocardial infarction Gastric ulcer Lobar pneumonia Staphylococcal septicaemia Septic arthritis
G G
C
Treatment
9 12
23 28
F M
I
S
Duration of disease Onset-death (yr)
Age at death
Sex
Patient
Multiple sclerosis Amyloidosis
Other conditions
5’
F
D. P. M. SYMMONS et al.
144
distribution in men was not inconsistent with that in the general population. In women there was a trend towards upper social class but it was not clear whether this was due to real effect or to an “up-grading” effect on the reporting of occupation on some death certificates [ 171. The medical record data were not sufficient to allow for a comparison of the groups in terms of therapy or smoking habits. Several aspects of the results suggest that there was in fact a real difference between men and women in their response to RA. First, although the relative risk of mortality was high for both sexes women showed more variation between referral groups and between intervals after presentation. Similar effects were found for excess mortality rates in women and, for both measures (RR and EMR), the effects were independent of age. These results suggest that initially women may have milder disease but with time this advantage is lost. The excess risks in men are distributed more uniformly over time. Second, the proportion of women dying from causes attributable to RA or its complications is strongly associated with age at presentation, the trends of increasing mortality with increasing age, being highly significant for both referral groups. The highest EMR (lO%/yr) is found in Group 2 women who were first seen after the age of 60 in whom the two effects-of age and duration-might be expected to have the greatest impact. Although the effect of age was also found in men in Group 1 the trend was much less marked. Thus, although RA may be more common in women, there may be a protective effect in younger women against its more severe complications. Thus, benefit may be hormonally mediated as it seems to be lost after the menopause. Recent reports suggest that users of oral contraceptives [18,19] may be protected from developing RA. Thus, there appears to be evidence for hormonal status not only influencing who may develop RA but also its ultimate prognosis. Although it might be considered that differences between men and women demonstrated here are due to the disparity of numbers (male:female ratio = 1: 1.8), the fact that the numbers of deaths observed and expected are comparable supports the conclusions of real differences between the sexes. These differences, particularly the more pronounced variation in risk in women, have implications for the validity of future long-term therapeutic studies in RA.
Acknowledgements-We thank the Endowment Fund of the Central Birmingham Health Authority, the Cancer Research Campaign and the Arthritis and Rheumatism Council for their financial support. We also acknowledge the help of staff at the National Health service Central Register for long-term tracing of patients.
REFERENCES I.
Cobb S, Anderson F, Bauer W: Length of life and cause of death in rheumatoid arthritis. N Engl J Med 249: 553-556, 1953 2. Duthie JJR, Brown PE, Truelove LH, Baragar FD, Lawrie AJ: Course and prognosis in rheumatoid arthritis: A further report. Ann Rheum Dis 23: 193-202, 1964 3. Uddin J, Kraus AS, Kelly HG: Survivorship and death in rheumatoid arthritis. Arth Rheum 13: 125-130, 1970 4. 5. 6. 7.
Monson RR, Hall AP: Mortality among arthritics. J Chron Dis 29: 459467, 1976 Allebeck P, Ahlbom A, Allander E: Increased mortality among persons with rheumatoid arthritis, but where RA does not appear on the death certificate. Stand J Rheum 10: 301-306, 1981 Allebeck P: Increased mortality in rheumatoid arthritis. Stand J Rheum 11: 81-86, 1982 Isomlki HA, Mutru 0, Koota K: Death rate and causes of death in patients with rheumatoid arthritis. &and J Rheum 4: 205-208,
8. 9.
1975
Koota K, Isomaki H, Mutru 0: Death rate and causes of death in rheumatoid arthritis patients during a period of five years. Sand J Rheum 6: 241-244, 1977 Prior P, Symmons DPM, Scott DL, Brown R, Hawkins CF: Cause of death in Rheumatoid Arthritis. Br J Rheum 23: 92-99,
1984
10. Prior P, Symmons DPM, Hawkins CF, Scott DL, Brown R: Cancer morbidity in Rheumatoid Arthritis. Ann Rheum Dis 43: 128-131, 1984 Office of Population Censuses and Surveys: The Registrar General’s Statistical Reviews for the years 1969-1973. London: HMSO 12. Office of Population Censuses and Surveys: The Registrar General’s Revised Estimates of the Population of England and Wales, 1961-71. London: HMSO, 1975 13. Ederer P, Axtell LM, Cutler SJ: National Cancer Institute (USA) Monograph No. 6, 1961 II.
Factors Influencing Mortality in RA
145
14. Peto R, Pike MC, Armitage NE ef al: Design and analysis of randomised clinical trials. II Analysis and examples. Br J Cancer 35: l-39, 1977 15. Rothman KJ, Boice, Jr JD: Epidemiological Analysis with a Programmable Calculator. Washington, USA: National Institutes of Health, 1979 16. Prior P: Cancer and rheumatoid arthritis: Epidemiological considerations. Am J Med 78(lA): 15-21, 1985 17. Prior P, Symmons DPM: Rheumatoid arthritis and social class. Br J Rheumatol 23: 309-310. 1984 18. Lingrave S, Kay CR. Reduction in incidence of rheumatoid arthritis associated with oral contraception. 19.
Lancet i: 569-57 I, 1978 Vandenbrouke JP, Valkenburg HA, Boerma JW et al. Oral contraceptives and rheumatoid arthritis. Further evidence for a preventive effect. Lancet ii: 839-842, 1982