Predictors of survival in systemic sclerosis Scleroderma.

403
PREDICTORS OF SURVIVAL IN SYSTEMIC SCLEROSIS
(SCLERODERMA)
ROY D. ALTMAN, THOMAS A . MEDSGER, JR., DANIEL A. BLOCH, and BEAT A. MICHEL
We conducted followup of 264 patients with
definite systemic sclerosis (SSc) who were entered into
the multicenter Scleroderma Criteria Cooperative Study
(SCCS) during 1973-1977. At the end of the study
(average 5.2 years of followup), 38% were known to be
alive, 50% were dead (68% of these deaths definitely
related to SSc), and 12% were lost to followup. Survival
analyses of 484 demographic, clinical, and laboratory
items recorded at entry into the SCCS (within 2 years of
physician diagnosis of SSc) were performed. Survival
declined linearly, and the cumulative survival rate was
<SO% at 2 years, 50% at 8.5 years, and 30% at 12 years
after entry. Analysis using combinations of entry variFrom the University of Miami School of Medicine and the
Arthritis Division, Miami Veterans Administration Medical Center,
Miami, Florida; the Division of Rheumatology and Clinical Immunology, University of Pittsburgh School of Medicine, Pittsburgh,
Pennsylvania; and the Divisions of Rheumatology and Biostatistics,
Stanford University School of Medicine, Stanford, California.
Supported in part by NIH grants AR-21393 and ZROl-GM21215-15, the United Scleroderma Foundation, the RGK Foundation, Austin, Texas, the American College of Rheumatology, the
Perlman Scleroderma Fund, Miami, Florida, the Arthritis Foundation, Western Pennsylvania Chapter (Shoemaker fund), and the
Miami Veterans Administration.
Roy D. Altman, MD: Professor of Medicine, University of
Miami School of Medicine, and Chief, Arthritis Division, Miami
Veterans Administration Medical Center; Thomas A. Medsger, Jr.,
MD: Professor of Medicine and Chief, Division of Rheumatology
and Clinical Immunology, University of Pittsburgh School of Medicine; Daniel A. Bloch, PhD: Senior Research Associate, Divisions
of Rheumatology and Biostatistics, Stanford University School of
Medicine; Beat A. Michel, MD: Fellow in Rheumatology, Stanford
University School of Medicine (current address: University Hospital of Zurich, Zurich, Switzerland).
Address reprint requests to Roy D. Altman, MD, Department of Medicine, University of Miami School of Medicine, PO Box
01696u (VA 111), Miami, F L 33101.
Submitted for publication January 23, 1990; accepted in
revised form October 12, 1990.
Arthritis and Rheumatism, Vol. 34, No. 4 (April 1991)
ables identifying organ system involvement confirmed
that renal, cardiac, pulmonary, and gastrointestinal
involvement in SSc predicted reduced survival; however, data on organ system involvement at study entry
could not be used to consistently predict which organ
system would ultimately be involved as the primary
cause of death. By survival tree analysis, the individual
entry variables best predicting reduced survival included older age (>64 years), reduced renal function
(blood urea nitrogen >16 mgldl), anemia (hemoglobin
5 1 1 gm/dl), reduced pulmonary diffusing capacity for
carbon monoxide (550% of predicted), reduced total
serum protein level ( 5 6 gm/dl), and reduced pulmonary
reserve (forced vital capacity <SO% with hemoglobin
>14 gmldl or forced vital capacity <65% with hemoglobin 1 1 4 gm/dl). Cox proportional hazards model
analysis confirmed these results. Different combinations
of variables led to markedly different survival rates. The
poorest prospects for survival were in patients with SSc
who were 1 6 4 years old with a hemoglobin level 511
gm/dl, and in those >64 years old with a blood urea
nitrogen level >16 mg/dl. These results may be useful in
predicting individual patients at risk for shortened
survival.
Systemic sclerosis (SSc; scleroderma) is classified as one of the connective tissue diseases. It is
characterized by vascular abnormalities such as
Raynaud’s phenomenon and telangiectasias; induration and thickening of the skin; and dysfunction of
other organs, including the kidneys, lungs, heart, and
gastrointestinal (GI) and musculoskeletal systems (13). Clinical severity and progression vary, from a
subtype of SSc with limited cutaneous changes to SSc
with diffuse cutaneous changes; the latter tends to be
ALTMAN ET AL
404
associated with more frequent and more severe visceral involvement (1,2).
M a n y retrospective studies on survival among
patients with SSc have been published (4-15). All of
the authors agree that the development of serious
visceral involvement in SSc portends early mortality.
T h e case series reported here is unique in that it
represents a geographically diverse, multicenter effort
in which patients were enrolled consecutively, mostly
within 2 years of diagnosis; comprehensive demographic, clinical, and laboratory data w e r e collected in
a standardized manner at entry; and many patients
w e r e followed for more than 10 years.
PATIENTS AND METHODS
Patients. Patients were identified through the multicenter Scleroderma Criteria Cooperative Study (SCCS),
performed by the Subcommittee for Scleroderma Criteria of
the American College of Rheumatology (formerly the American Rheumatism Association) Diagnostic and Therapeutic
Criteria Committee. The primary aim of this investigation
was to develop preliminary criteria for the classification of
SSc (16). The methods for data collection, patient entry, and
diagnosis verification procedures, as well as entry demographic, clinical, and laboratory features of the patients have
been reported (16,17). Briefly, between 1973 and 1977, 29
centers referred 264 patients with recently diagnosed definite
SSc. Almost all were entered within 2 years of first being
diagnosed by a physician as having SSc. A detailed protocol
containing 484 items of information was completed on each
patient, and each had a serum specimen stored. Followup
data were requested 1 year after entry to verify diagnosis,
and a 3-physician review panel examined all forms to confirm the entry diagnosis.
In a second followup conducted at least 3 years after
entry, the diagnosis was reconfirmed, patient outcome was
determined, and information on 133 clinical and laboratory
items was obtained. Sources of information were the patient,
center coordinator, patient’s primary physician (if different
from the center coordinator), and chart review. A final
followup was conducted during 1983-1985. Patients who
could not be readily located were traced through relatives,
initial referring physicians, hospital records, telephone
books, state motor vehicle registration systems, the Social
Security Administration, and the National Death Index
(National Center for Health Statistics, US Department of
Health and Human Services, Hyattsville, MD).
For patients known to have died, physician’s office,
hospital, autopsy, and death certificate records were reviewed to establish the date and cause of death. Each death
was categorized as either definitely SSc-related or possibly
SSc-related. A patient was considered to have a definitely
SSc-related death only if there was a clear medical association with SSc. For example, a death due to end-stage
diabetes mellitus, malignancy, or suicide was not assigned to
the definitely SSc-related category. If a patient died of
septicemia in the setting of SSc renal crisis, the death was
assigned to the definitely SSc-related category. Deaths attributed to SSc were further subdivided based on the organ
system primarily responsible for the fatality, as above.
During the past decade, several serum autoantibody
systems that are relatively specific for SSc and are believed
to have important clinical subclassification and prognostic
significance have been identified (18,19). Sera saved frozen
at -70°C from the original study were thawed, and anticentromere and anti-Scl-70 (anti-topoisomerase I) antibody
tests were performed for inclusion in the present analysis.
All data were entered into the data bank of the
Arthritis, Rheumatism, and Aging Medical Information System. Computer printouts were reviewed for accuracy and
internal consistency. Twenty percent of the printouts were
compared with the original data collection forms; the percentage of entry errors was 0.05% in that sample.
Definitions of organ system involvement. Organ system involvement at the initial visit was defined using the
criteria published by Medsger et a1 (lo), with modifications.
Only clinically detected antemortem evidence of involvement was used, and findings were accepted only when not
attributable to some other disease process (e.g., congestive
heart failure due to atherosclerotic heart disease). The
findings listed below were required in order to classify a
patient as having involvement of the particular organ system.
Gastrointestinal involvement was defined as distal
esophageal hypo- or aperistalsis, documented by either
cineradiographic or manometric studies, typical small bowel
malabsorption syndrome, or colonic “sacculations” characteristic of scleroderma. Pulmonary involvement was diagnosed by any one of the following: bilateral basilar interstitial fibrosis or infiltration on chest radiograph, active
pleuritis (pain plus friction rub), cardiac catheterizationproven pulmonary hypertension, or diffusing capacity for
carbon monoxide (DLco) reduced to <13 ml/minute/mm Hg.
Cardiac involvement was characterized by any one of the
following: active pericarditis (pain plus friction rub), congestive heart failure, or nodal or ventricular arrhythmias. We
defined scleroderma renal involvement as rapid, progressive
renal failure or proteinuria of 23.5 gmlday. Patients who did
not meet any of the above criteria were classified as having
no major organ system involvement.
Statistical analysis. Demographic, clinical, and laboratory variables used in the analysis were derived from the
initial SCCS entry visit data. The predictive value of each of
the 484 variables with regard to association with death was
initially examined by univariate statistical methods for differences in means (continuous data) or frequencies (categorical data), for all patients who died versus all who survived.
Since this approach does not account for the factor of time,
a second univariate analysis compared 2 patient groups with
respect to time to death: those who died within 2 years of
entry versus those who were alive at the end of the study and
had been followed more than 2 years from entry. No
variables were lost, and a few additional variables were
found to be significant, using this approach. Variables significant at the designated P < 0.05 level, as well as other
variables considered important based on previous investigations (4-15), were chosen for multivariate study. Table 1 lists
the 72 selected variables, subdivided into 9 subsets.
SURVIVAL IN SSc
Multivariate analysis included creation of a survival
tree (20,21). With this method, a group of patients is split
into two subgroups according to values for a selected variable. A split partitions the group into patients with values
lower than a particular cutoff point and those with values
greater than this cutoff point, e.g., age 564 years and age
>64 years. At every group of the tree which is to be
subdivided, all variables are examined, and the “best” one
is selected, along with its splitting value. Best is assessed in
terms of a goodness-of-splitindex that can be evaluated for
any split of any group of the tree. In this study, the
Mantel-Haenszel (log rank) statistic was used. Two subgroups result from every potential split, and these are treated
as the two samples for which this statistic is computed. The
split implemented at any group of the tree is that which
results in the highest value of the Mantel-Haenszelstatistic.
As defined, the (best) split selected most clearly separates
the two derived subgroups’ Kaplan-Meier survival curves,
which were used to estimate survival (22).
Multivariate analysis with the survival tree method
and the stepwise Cox proportional hazards regression model
(22) was performed separately on each of the 9 subsets,
containing 5-1 1 variables each, as listed in Table I. The Cox
proportional hazards regression model was also applied to all
72 variables in these subsets. This model does not allow for
missing values; hence, mean normal values were entered as
replacements, on the premise that a missing value for a
particular variable was most likely to be normal. Many of the
72 variables had some values missing, and a few had a high
proportion of missing values (e.g., erythrocyte sedimentation rate [ESR] 35%, Po, 50%, and DLco 29% of predicted)
(Table 1).
A survival curve was constructed for a general
population that was matched to the patient group for age at
entry, year of entry, sex, race, and length of followup. This
was based on survival probabilities as published by the
National Center for Health Statistics. Organ system involvement was related to survival, using Kaplan-Meier survival
curves and Mantel-Haenszel statistics.
RESULTS
The mean (kSEM) age at the time of study
entry was 49.1 +- 0.8 years. Study entry took place an
average of 5.7 _t 0.2 years after appearance of the first
symptom attributable to SSc and 1.9 t 0.1 years from
the time of physician diagnosis. Followup averaged 5.2
t 0.2 years from entry and 7.2 -I 0.2 years from
diagnosis.
The survival curve (Kaplan-Meier) for all 264
patients from the date of entry to the last followup is
shown in Figure 1. The survival curve for the general
population (matched to the SSc group) is also displayed in Figure 1 . At 2 years, survival in the general
population was >98%; survival in the SSc group was
<80%. At the end of 12 years, survival in the general
405
population was 89%, but the survival rate was only
30% for the SSc patients.
Of the 264 patients, 101 (38%) were known to be
alive at the last followup, 131 (50%) were known to
have died, and 32 (12%) were lost to followup. In this
group of 264 patients with SSc, all patients had generalized disease, with arm, face, and/or truncal involvement. Although 63 patients could be identified as
having CREST syndrome (calcinosis, Raynaud’s phenomenon, esophageal dysmotility, sclerodactyly, and
telangiectasias), REST, or CRST, they also appeared
to have diffuse cutaneous disease.
Eighty-nine (68%) of the 131 deaths were definitely related to SSc, and 42 (32%) were possibly
related to SSc (Table 2). Of the deaths attributed to
SSc, renal involvement predominated as the cause of
death (39%). In 2 patients, it was not possible to
separate cardiac from pulmonary SSc as the event
leading to death. Of the deaths possibly related to SSc,
cardiac disease was most frequent (36%), especially
atherosclerotic cardiovascular disease with myocardial infarction. There were 18 cancers detected in 16
(6%) of the 264 patients, all within 4 years of the
diagnosis of scleroderma, and 9 deaths were attributed
to malignancy. The primary cancer sites included lung
(4 patients), breast (3 patients), leukemia (3 patients),
colon (2 patients), skin (2 patients), carcinoid (1 patient), cervix (1 patient), common bile duct (1 patient),
and kidney (1 patient).
Findings of univariate analysis. Univariate analysis was performed for all variables, comparing the 58
patients who died within 2 years with the 94 patients
who were alive at the end of the study and were
followed longer than 2 years from the time of entry
into the SCCS. The statistical significance of the
differences between the two groups are presented in
Table 1. Among patients who were taking prednisone,
the mean daily dosage was higber, but not statistically
significantly higher, in those who died in less than 2
years (mean ? SD 29 If: 6 mg) than in those who were
alive at last followup and were followed longer than 2
years (19 t 4 mg) (P< 0.08). Corticosteroids were the
only therapeutic agents for which there were adequate
data for analysis. Microscopic hematuria, when
present, was greater in those who died in less than 2
years (mean SEM 10 ? 5 red blood cells/high power
field) than in those who were followed longer than 2
years and alive at last followup (2 ? 0.5 red blood
celldhigh power field) (P < 0.001).
Organ system involvement. Kaplan-Meier survival curves according to organ system SSc involve-
*
ALTMAN ET A L
406
Table 1. Univariate analysis of 72 variables: differences between patients who died within 2 years of
followup and those who were alive at the end of the study and were followedup more than 2 years*
Variable
General
Age at entry, years
Sex (% female)
Race (% nonwhite)
Married (%)
Physical laborer (%)
Tobacco use, pack years
Alcohol use, index
Pulse, beats per minute
Microangiopathy by blood smear (%)
Taking prednisone (%)
Dermatologic
Dry eyes (%)
Alopecia (%)
Generalized scleroderma (%)
Interlip distance, m m
Digital tip pitting scars (%)
Musculoskeletal
Promixal muscle weakness (%)
Symmetry of joint involvement (%)
Tenosynovitis (%)
Joint swelling (%)
Joint deformity (%)
CPKt
Serum aldo1ase.t
Gastrointestinal
Dysphagia (%)
Fecal fat excretion, 72 hours, gmiday
D-xylose test, gm
Abnormal esophageal manometry (%)
Colonic sacculations (%)§
P u 1m on a r y
Dyspnea (%)
Dry basilar rales (%)
Increased cardiac pulmonic second
sound (P2) (%)
Bibasilar interstitial fibrosis (%)§
Pleural effusion (%)§
Arterial Po,, m m Hg
Vital capacity, ml
Forced vital capacity, % predicted
Diffusion capacity, mm Hgiminute
Diffusion capacity, % predicted
Cardiac
Orthopnea (%)
Neck vein distention (%)
Ventricular gallop (%)
Pedal edema (%)
EKG
Right ventricular hyperthrophy (%)
Nonspecific ST-T changes (%)
Arrhythmia (%)
Left ventricular enlargement (%)$
Renal
Blood pressure, m m Hg
Systolic
Diastolic
BUN, mg/dl
Serum creatinine, mg/dl
Proteinuria, I + to 4+ (%)
Dead within
2 years
(n = 58)
Alive and
followedup
>2 years
(n = 94)
Pt
*
46
<0.01
*
22 f 3
28 f 8
81 t 1
0
13
12
7
97
53 2
28
24
9
5
47 t 7
60 t 22
88 I
6
31
f
1
19
14
8
12
Values
entered
(%)
NS
<0.001
NS
100
I00
100
99
97
92
89
100
60
<0.01
100
<0.05
<0.05
100
<0.001
NS
NS
NS
NS
2
17
100
42 f 1
53
45 t I
43
<0.001
100
43
19
100
28
67
1.7 f 0.3
1.0 f 0.1
15
9
I00
14
45
1.0 ? 0.3
0.7 t 0.1
<0.001
NS
NS
<0.05
<0.05
NS
CO.05
100
100
22
99
86
84
39
12
6.1 t 1
3.1 ? 0.5
76
17
5
5.1 t 1
5.4 f 0.3
62
14
NS
NS
80
11
29
100
47
NS
NS
<0.001
<0.01
NS
100
100
82
67
31
21
36
13
9
<0.001
<0.05
I00
CO.01
100
37
14
74.6 f 2.0
2,523 f 133
69 t 3.2
14 t 1
64 t 5
14
1
<0.01
98
99
50
84
82
73
71
87.2 f 1.5
3,100 f 88
84 f 1.7
18 t 1
83 +- 3
<0.01
<0.001
<0.01
<0.001
<0.01
<0.01
19
19
22
26
3
1
3
4
<0.001
<0.01
14
34
31
26
2
23
13
4
<0.001
<0.01
132 f 3
81 f 2
25 f 5
1.2 f 0.1
35
124 ? 2
78 2 1
14 2 1
0.9 f 0.03
10
<0.01
<0.001
<0.01
<0.001
<0.05
NS
<0.05
<0.05
<0.001
100
99
78
81
99
97
95
92
96
100
100
96
98
95
SURVIVAL IN SSc
407
Table 1. (Cont’d)
Variable
Renal
Urine
>O RBC/HPF (%)
>O granular casts/HPF (%)
Creatinine, mg/24 hours
Creatinine clearance, m h i n u t e
Immunologic
Positive skin test for mumps (%)
Serum globulin, gm/dl
Rheumatoid factor, latex
agglutination titer
Log latex fixation, titer
Fluorescent antinuclear antibody titer
Anticentromere antibody (%)
Anti-topoisomerase I (%)
Laboratory
Hemoglobin, g d d l
WBC, x I ,000/mm3
% neutrophils
% lymphocytes
% monocytes
% eosinophils
Westergren ESR, mm/hour
Total protein, gm/dl
Albumin, gmidl
SGOT$
LDHt
Dead within
2 years
(n = 58)
35
21
946 2 65
84 t 6
Alive and
followedup
<2 years
(n = 94)
Pt
(%I
20
4
1,073 t 42
88 t 3
<0.01
<0.01
NS
NS
96
96
17
73
Values
entered
46
3.2 t 0.1
2,429 2 1,785
73
3.2 2 0.1
302 t 72
<0.05
NS
NS
51
92
86
8.2 t 2.3
4,274 f 3,844
5
51
13.4 t 0.7
162 t 35
19
21
<0.001
0.05
<0.05
<0.01
86
88
78
76
12.3 t 0.3
9.0 2 0.5
71 f 2
21 t 1
5 2 1
2 t 1
42 t 5
6.8 t 0.1
3.5 f 0.1
1.0 t 0.2
1.3 t 0.1
13.2 -t 0.2
1.3 t 0.2
67 f 1
24 t 1
6-Cl
3 t l
23 t 3
7.0 t 0.1
3.8 t 0.1
0.7 t 0.1
1.0 t 0.1
<0.05
97
99
97
97
97
89
65
93
93
97
17
<0.01
<0.05
<0.05
<0.05
<0.05
<0.01
NS
<0.001
<0.05
<0.01
* Unless otherwise indicated, values are the mean f SEM. CPK = creatine phosphokinase; EKG =
electrocardiography; BUN = blood urea nitrogen; RBC = red blood cells; HPF = high power field;
WBC = white blood cells; ESR = erythrocyte sedimentation rate; SGOT = serum glutamic
oxaloacetic transaminase; LDH = lactate dehydrogenase.
t The significance of the difference in means was assessed with the 2-sample t-test. The significance
of the difference in percents was assessed with the chi-square test for 2 x 2 tables. NS = not
significant.
$ Multiple of normal, derived by dividing the observed value by the upper limit of normal value for the
laboratory.
J By radiography or barium contrast radiography.
ment at entry into the study are displayed in Figure 2.
Renal disease had the poorest median (_tSE) survival,
at 3 t 25 months (i.e., 50% of the group had died at 3
months), and 7 of 10 patients (70%) did not survive
until the end of the study. Overall survival was poorer
for patients with reral disease than that for patients
with GI involvement (with no renal, cardiac, or pulmonary disease) (P = 0.03, by Mantel-Haenszel test)
or with no major organ system involvement ( P = 0.02).
Cardiac involvement (without renal involvement) predicted a median survival time of 32 5 40
months, and 9 of 15 patients did not survive until the
end of the study. Patients with pulmonary SSc (with
no renal or cardiac involvement) survived a median of
78 f 17 months, but 60 of 104 (58%) died before the
last followup. These patients with pulmonary SSc had
significantly poorer survival prospects than did patients with GI involvement with no renal, cardiac, or
pulmonary disease ( P = 0.04, by Mantel-Haenszel
test). Patients with GI involvement (with no renal,
cardiac, or pulmonary disease) survived a median of
99 17 months, and 35 of 84 (42%)died before the last
followup. SSc patients with no renal, cardiac, pulmonary, or GI involvement survived for a median of 108
2 9 months, and 20 of 51 (39%) did not survive until
the end of the study. Other group differences with
respect to survival were not statistically significant
(Mantel-Haenszel test).
Of the 10 patients in whom renal involvement
had been identified at the time of study entry, 6 died of
renal disease and 1 died of cancer. However, in 29 of
the 35 patients who died of SSc renal disease (83%),
*
ALTMAN ET AL
408
Table 2. Diseased organ system or event leading to death in
patients with systemic sclerosis (SSc)
0
0
24
48
72
96
120
144
168
MONTHS OF FOLLOW-UP
Cause of death
Definitely
related to
ssc
(n = 89)
Renal
Cardiac
Pulmonary
Gastrointestinal
Cancer
Cardiopulmonary
Infection
Suicide
Vasculitis
Cerebrovascular accident
Musculoskeletal
Other*
35
18
18
13
0
2
0
0
2
0
1
0
Possibly
related
to ssc
(n = 42)
Total
(n = 131)
35
33
18
14
9
2
3
3
2
2
0
15
0
1
9
0
3
3
0
2
0
9
1
9
Figure 1. Observed wrvival curve for 264 patients with systemic
sclerosis (SSc) (0)and expected survival for a general population
matched to the SSc group for date of entry into the study, duration
of followup, age, sex, and race (E).
* One patient died of complications of second- and third-degree
burns over 80% of the body; 8 died of unknown causes.
there was no evidence of renal involvement at the time
of the initial evaluation. In general, there was little
direct correlation between organ system involvement
at the time of the initial examination and recorded
cause of death, except in the case of renal involve-
ment. Both loose and strict definitions of organ system
involvement yielded similar low correlations.
Findings of multivariate analysis. Variables
found to be significant by univariate analysis and those
reported in the literature as predictive of survival were
"t,
0
24
I
48
I
72
I
96
I
MONTHS OF FOLLOW-UP
120
I
--
1
L..
..."
-
144
:
"I
168
1
Figure 2. Kaplan-Meier survival curves for 264 patients with systemic sclerosis (SSc), by organ system involvement at
entry into the study. Curves are shown for those with renal involvement (R; n = lo), heart involvement without renal
involvement (H; n = 15), pulmonary involvement without renal or heart involvement (P; n = 104), gastrointestinal
involvement without renal, heart, or pulmonary involvement (G; n = 84), and SSc without renal, heart, pulmonary, or
gastrointestinal involvement (N; n = 51).
409
SURVIVAL IN SSc
subjected to multivariate analysis, using a survival tree
and a stepwise Cox proportional hazards model. Figure 3 depicts the survival tree that resulted from an
analysis in which the 72 variables from Table 1 were
included as being potentially predictive. The strongest
predictor of survival was age at study entry. The
median survival for the 32 patients who were >64
years old at entry was 25 months (2.1 years), whereas
that for the 232 patients who were 5 6 4 years old at
entry was 107 months (8.9 years). The combinations of
Table 3. Terminal subgroups from survival tree
Median survival
in months*
Patient characteristicst
8 (3)
Age 564 years
Hb 5 1 1 gmidl
9 (6)
Age >64 years
BUN >16 mgidl
30 (7)
Age 564 years
Hb > 1 1 gm/dl
DLco 550% predicted
35 (9)
Age 564 years
Hb > 1 1 gm/dl
DLco >50% predicted
Total serum protein 5 6 gmidl
45 ( 5 )
Age >64 years
BUN 5 1 6 mg/dl
46 (15)
Age 564 years
Hb >14 gmidl
DLco >50% predicted
Total serum protein >6 gm/dl
FVC <80% predicted
89 (13)
Age 564 years
Hb > 1 1 gm/dl and 5 14 gm/dl
DLco >50% predicted
Total serum protein >6 gmidl
FVC 5 6 5 % predicted
104 (16)
Age 564 years
Hb >14 gm/dl
DLco >50% predicted
Total serum protein >6 gmidl
FVC 280% predicted
153+ (14)
Age 564 years
Hb > 1 1 gm/dl and 5 14 gmidl
DLco >50% predicted
Total serum protein >6 gmidl
FVC >65% predicted
* Numbers in parentheses are the terminal subgroup numbers
shown in Figure 3.
I' Hb = hemoglobin; BUN
= blood urea nitrogen; DLco = diffusing
capacity for carbon monoxide; FVC = forced vital capacity.
Figure 3. Survival tree for 264 patients with systemic sclerosis,
using all of the variables shown in Table 1. The tree is derived from
recursive partitioning (see Patients and Methods). Large circles
show the variable on which the tree divides. The final subgroups
formed are represented by squares. The upper number within each
large circle and each square is the number of patients; the lower
number (in parentheses) is the median survival, in months. Small
circles identify the number of the terminal subgroup in the tree.
Values at the arms of the tree represent the value upon which the
variable divides. Hb = hemoglobin; BUN = blood urea nitrogen;
DLCO = diffusing capacity for carbon monoxide; FVC = forced
vital capacity.
variables from the terminal subgroups from the survival tree are also displayed in Table 3 .
Entry characteristics of survival tree subgroups
did not consistently predict which initially involved
organ system or event would lead to death. Azotemia
was present in patients who died of renal disease, but
did not consistently predict renal disease; it was often
present in patients who died of cardiac involvement or
cancer.
Kaplan-Meier survival curves (Figure 4) depict
the differences in survival between the various terminal groups, e.g., the curves for subgroups 4 and 3
(Figure 4A) illustrate the association of reduced survival with hemoglobin levels I1 1 gm/dl in SSc patients
ALTMAN ET AL
410
A
0
I
0
24
I
io
!
48
72
96
120
MONTHS OF FOLLOW-UP
I
I
144
168
B
040
I
24
48
72
96
120
MONTHS OF FOLLOW-UP
144
168
C
80
>
Ob
0
I
24
48
72
96
120
MONTHS OF FOLLOW-UP
144
168
Figure 4. Survival curves for the 264 patients with systemic sclerosis, by subgroups (circled numbers at the end of each curve) from
Figure 3 and Table 3. Curves are shown for the group with the
poorest survival (A), the group with the median survival (B),and the
group with the best survival (C).
age 564 years. In contrast, SSc patients classified in
subgroup 14 (Figure 4C) had the most favorable prospects for survival. Subgroup 14 was further split with
regard to the variables of white blood cell count
(number of cells/ml) and urine creatinine level (mg/24
hours). These splits are not shown because they were
relatively weak and increased the complexity of the
tree structure.
A stepwise Cox proportional hazards analysis
was performed on the 72 variables from Table 1
(results not shown). Most of the variables selected by
the survival tree method were high in the stepwise COX
model. There were 6 variables that did not appear in
the survival tree analysis but appeared to be significant
( P < 0.05) in the stepwise Cox model. These were
proximal muscle weakness, absolute lymphocytosis,
proteinuria, microscopic hematuria, reduced arterial
Po,, and radiographic evidence of left ventricular
enlargement. Although individually these variables
were predictive of reduced survival, cross correlations
indicated interdependence with the selected tree variables. For example, proximal muscle weakness was
significantly correlated with abnormal lung function
(low DLco [% of predicted] and low forced vital
capacity [FVC; % of predicted]), azotemia, and/or
anemia.
The effects of individual variables can be quantified using the relative risks estimated from a stepwise
Cox model (Table 4).As assessed by the Cox model,
all variables shown in the tree had significant associations with death. Thus, the negative coefficient for
FVC (Table 4) implies that patients with lower FVC
have poorer survival prospects. The relative risk of
0.98 indicates that, with all else being equal, a patient
with a 1% higher FVC has a 2% reduced hazard of
death. In contrast, with the tree method, 4 subgroups
were defined (in terms of characteristics at entry) that
have distinct survival prospects that depend on the
FVC as well as other variables (subgroups 13-16,
Table 3). The two methods pursue different goals: the
Cox model ascribes an effect to a variable, while the
tree defines subgroups that have distinct survival
prospects. Hence, it was not surprising when variables
found to be predictive in the tree structure, such as
blood urea nitrogen (BUN) level, hemoglobin level,
white blood cell count, and 24-hour urine creatinine
level, were not significant in the Cox proportional
hazards model when all variables were included.
A tree structure as well as a Cox proportional
hazards model was developed for each of the 9 subsets
listed in Table 1 (not shown). The findings from these
41 1
SURVIVAL IN SSc
Table 4.
Stepwise Cox proportional hazards model applied to survival tree variables (including sex)
~~
Step
no.
Variable*
1
2
3
4
5
6
7
8
FVC (% predicted)
BUN (mg/dl)
Hemoglobin (gm/dl)
WBC (x1,000/mm3)
DLco (% predicted)
Total serum protein (gmldl)
Age (years)
Sex (M/F)
Coefficient
rf-
SE
-0.018 rf- 0.006
0.016 rf- 0.004
-0.162 f 0.052
0.139 f 0.038
-0.013 2 0.004
-0.325 2 0.128
0.018 rf- 0.008
0.387 ? 0.216
Relative
risk?
Improvement
0.98
1.02
0.85
1.15
0.99
0.72
I .02
1.47
<0.001
<0.001
0.003
<0.001
0.012
P
0.005
0.016
0.122
* FVC = forced vital capacity; BUN = blood urea nitrogen; WBC = white blood cells; DLco =
diffusing capacity for carbon monoxide.
-F Relative risk = relative multiplicative effect of the variable on the hazard function corresponding to
a I-unit increase in that variable only.
subset analyses added little insight to the more inclusive multivariate analyses already described, which
considered all 72 variables as potentially predictive of
shortened (or lengthened) survival. To illustrate the
results, both multivariate methods found FVC and
DLco to be important discriminators in the pulmonary
subset, BUN in the renal subset, and neck vein and
left ventricular enlargement by radiography in the
cardiac subset. Further details are available from the
authors upon request.
Prior investigations (12,23,24) have found male
sex to be a significant predictor of poor survival. In the
present study, however, sex was not a predictor of
outcome by either univariate or multivariate analysis
(Table 4).
DISCUSSION
Systemic sclerosis has been reported to have an
annual incidence of 1-20 new cases per million population (1,25), a prevalence of 1.3-10.5 per 100,000
population (14,25), and an annual mortality rate of 1.8
per million population (14). Overall mortality has been
estimated at 39-65% (6,9,10,12,13,15). It has been
suggested that these prior reports of SSc mortality are
biased by including a larger proportion of the more
seriously ill patients. Other reports have indicated a
lower annual incidence of disease (13,26) but concur
with the mortality rate of near 50%.
In this study, we evaluated 264 patients with
definite SSc, who were referred over a decade ago
from 29 centers in North America, for development of
criteria for the classification of the disease (16). The
severity of SSc is apparent in that 50% of the patients
are known to have died during a followup period that
averaged 7.2 years from diagnosis and 5.2 years from
entry into this study. The failure of the survival rate to
“level off’ and parallel the expected survival for the
general population suggests that the microvascular and
fibrotic changes of SSc importantly reduce the reserve
function of various organ systems. Thus, SSc patients
would be at greater risk for mortality from other
superimposed illnesses, such as degenerative vascular
diseases and pneumonia. The dramatic reduction in
survival of the entire SCCS group is emphasized when
the group is compared with a matched population
(Figure l), and is reminiscent of the curves expected
with many forms of cancer.
Most earlier studies have utilized univariate
analysis to determine predictors of outcome. There are
similarities, but also differences, between our findings
and those in the prior reports. Older age is a consistent
predictor of poor survival. Poorer survival has been
predicted for patients age >40 (5,9), age >45 (lo), and
age >50 (5,8,12,15). In the present study, age >64
years was selected as the main characteristic associated with poor prognosis. Poor prognosis had also
been reported for nonwhites (12), women (12,23,24),
physical laborers (13), and married persons (12-14),
but these factors were not significantly associated with
death in our study, by either univariate or multivariate
methods. Similarly, poor prognosis previously reported in association with cigarette smoking (12,14),
presence of a ventricular gallop (15), use of corticosteroids (15), elevated ESR, cardiac arrhythmia (27), and
abnormal changes on electrocardiography (9) were
shown to be present in this study by univariate techniques, but not after accounting for covariates.
Univariate and multivariate analyses support
the reported poorer prognosis related to reduced hemoglobin (5,12), elevated BUN level (9), proteinuria
(12), reduced DLco (23), and reduced FVC (23).
Although predictive by univariate analysis, multivariate analysis did not demonstrate any predictive power
412
for anticentromere or anti-Scl-70 antibody. The latter
has been found to be both predictive (28) and not
predictive (19) of survival, in separate studies.
In contrast to two previous reports (5,12), average hemoglobin levels among the patients we studied
were very similar for those who survived (mean 2
SEM 13.2 ? 0.2 gm/dl) and those who died during the
study (12.3 ? 0.3 gm/dl). However, as we found when
using the survival tree analysis, the association of
hemoglobin is non-monotonic: Both low (anemia) and
high (polycythemia, perhaps chronic lung disease)
hemoglobin levels are poor prognostic features. Such a
distinction was not detected using the Cox proportional hazards model analysis, representing an advantage to the survival tree method.
Prior studies have emphasized involvement of
certain organ systems as predicting poor prognosis,
including diffuse skin disease (5,29), truncal skin involvement (6,9,28), pulmonary disease (9,10,12), cardiac disease (4,5,10,12), and renal disease (4,5,7,8,
10,12). Among the patients in this study, the frequency
of truncal skin thickening at entry was nearly the same
in survivors (44%) and nonsurvivors (48%). Considering the findings in other studies, this result was unexpected, but it may reflect the fact that the majority of
patients in this study developed diffuse cutaneous
involvement, even though at the time of entry, truncal
skin thickening was not recorded. The most common
SSc-related cause of death was renal involvement,
emphasizing the severity of this complication. The
majority of the SCCS patients died prior to the availability of effective antihypertensive therapy, such as
angiotensin-converting enzyme inhibiting agents.
Since effective therapy is dependent upon early recognition of “renal crisis,” it is most important to identify
the patients at greatest risk.
Survival curves for the subsets derived from the
survival tree reveal combinations of features that
predict increased rates of early and late mortality. For
example, in patients with SSc who were 5 6 4 years
old, anemia (hemoglobin I1 1 gm/dl) predicted a median survival of only 8 months (subset 3, Figure 3)
while a higher hemoglobin level (> 11 gm/dl) predicted
a median survival of 126 months (subset 4). Also, in
patients >64 years old, a BUN level >16 mg/dl (subset
6) predicted a median survival of 9 months, versus 45
months in those >64 years old with a BUN level 5 1 6
mg/dl (subset 5). Similar conclusions emerge with
regard to total serum protein (subsets 9 and 10).
Despite the inconsistencies found when attempting to relate subgroups to organ system or event
ALTMAN ET AL
leading to death, there were some trends within the
subgroups. For example, patients in subgroup 3, with
the worst median survival, had anemia and severe
involvement of the kidney. Subgroup 6, with the
second shortest median survival time, comprised older
individuals with involvement of the lungs and severe
involvement of the heart and kidneys. Patients in
subgroup 7, with the next worst outcome (median
survival 30 months), had significant lung involvement.
These relative priorities (kidney, heart, lung) are consistent with the findings of other studies (10). Features
in other subgroups suggest additional risk factors, i.e.,
malabsorption (subgroup 9), older age (subgroup 5 ) ,
and lung disease (subgroups 13 and 15). In general,
these findings are supported by the proportion of
patients who died of the associated causes.
Although not as appealing clinically, the use of
combinations of individual variables, some of which
represent disease in more than one organ system, has
greater predictive value than overall use of organ
system involvement. It is suggested that the information obtained from the Cox proportional hazards
model and the survival tree is more useful than organ
system involvement data for describing patient outcome. This is emphasized by the nonspecific indicators of poor health detected using the survival tree,
e.g., older age, low total serum protein levels, abnormal pulmonary function, both high and low hemoglobin levels, and high and low BUN values. Proximal
muscle weakness was increased in all 4 of the worst
prognosis subgroups (Figure 3), and increased urinary
protein was present in 3 of the 6 survival tree subgroups with the worst prognosis. Although not as
sensitive as the tree variables presented, reduced
peripheral white blood cell count and reduced urinary
creatinine also predicted reduced survival.
In summary, we have studied a large number of
patients with diffuse cutaneous systemic sclerosis over
a prolonged period of time and identified several entry
features associated with reduced survival. We are not
aware of other studies that have followed as large a
group of patients for a similar length of time or have
utilized such a variety of analytic techniques. The
findings reported herein may be of value in identifying
specific SSc patients who are at risk for shortened
survival.
ACKNOWLEDGMENTS
We acknowledge the late Dr. Gerald P. Rodnan,
whose guiding force created the data bank used in this study.
SURVIVAL IN SSc
We also acknowledge Dr. Alfonse T. Masi, who contributed
to the design of the data collection protocol, and the many
contributors to the SCCS. It is impossible to individually
recognize the multiple contributors, including statistics personnel, outcome assessors, and stenographers.
REFERENCES
1 . Medsger TA Jr: Systemic sclerosis (scleroderma), eosinophilic fasciitis, and calcinosis, Arthritis and Allied
Conditions. Eleventh edition. Edited by DJ McCarty.
Philadelphia, Lea & Febiger, 1989
2. Seibold JR: Scleroderma, Textbook of Rheumatology.
Third edition. Edited by WN Kelley, ED Harris Jr, S
Ruddy, CM Sledge. Philadelphia, WB Saunders, 1989
3. Rocco VK, Hurd ER: Scleroderma and scleroderma-like
disorders. Semin Arthritis Rheum 16:22-69, 1986
4. Lally EV, Jimenez SA, Kaplan SR: Progressive systemic sclerosis: mode of presentation, rapidly progressive disease course and mortality based on an analysis of
91 patients. Semin Arthritis Rheum l8:1-13, 1988
5. Farmer RG, Gifford RW, Hines EA Jr: Prognostic
significance of Raynaud’s phenomenon and other clinical characteristics of systemic scleroderma. Circulation
21:1088-1095, 1960
6. Tuffanelli DL, Winkelmann RK: Systemic scleroderma:
a clinical study of 727 cases. Arch Dermatol84:359-371,
1961
7. Medsger TA Jr: Epidemiology of progressive systemic
sclerosis, Current Topics in Rheumatology: Systemic
Sclerosis (scleroderma). Edited by CM Black, AR Myers. New York, Cower Medical Publishing Ltd, 1985
8. Masi AT, D’Angelo WA: Epidemiology of fatal systemic
sclerosis (diffuse scleroderma). Ann Intern Med 66:87&
883, 1967
9. Bennett R, Bluestone R, Holt PJL, Bywaters EGL:
Survival in scleroderma. Ann Rheum Dis 30:581-588,
1971
10. Medsger TA Jr, Rodnan GP, Benedek TG, Robinson H:
Survival with systemic sclerosis (scleroderma): a lifetable analysis of clinical and demographic factors in 309
patients. Ann Intern Med 75:369-376, 1971
11. Medsger TA Jr, Masi AT: Epidemiology of systemic
sclerosis (scleroderma). Ann Intern Med 74:714-721,
1971
12. Medsger TA Jr, Masi AT: Survival with scleroderma. 11.
A life-table analysis of clinical and demographic factors
in 358 male U.S. veteran patients. J Chronic Dis 26:647660, 1973
13. Medsger TA Jr, Masi AT: The epidemiology of systemic
sclerosis (scleroderma) among male U.S. veterans. J
Chronic Dis 31:73-85, 1978
14. Neuberger JS, Neuberger GB: Epidemiology of the
rheumatic diseases. Nurs Clin North Am 19:713-725,
1984
15. Wynn J, Fineberg N, Matzer L, Cortada X, Armstrong
413
W, Dillon JC, Kinney EL: Prediction of survival in
progressive systemic sclerosis by multivariate analysis
of clinical features. Am Heart J 110:123-127, 1985
16. Subcommittee for Scleroderma Criteria of the American
Rheumatism Association Diagnostic and Therapeutic
Criteria Committee: Preliminary criteria for the classification of systemic sclerosis (scleroderma). Arthritis
Rheum 23581-590, 1980
17. Masi AT, Medsger AT Jr, Altman RD, Rodnan GP,
Fries JF, Brown BW, D’Angelo WA, LeRoy EC, Mackenzie AH, McShane DJ, Sharp GC: Methods and preliminary criteria for the classification of systemic sclerosis (scleroderma). Clin Rheum Dis 5:2748, 1979
18. Tan EM, Rodnan GP, Garcia I, Moroi Y, Fritzler MJ,
Peebles C: Diversity of antinuclear antibodies in progressive systemic sclerosis: anti-centromere antibody
and its relationship to CREST syndrome. Arthritis
Rheum 23:617-625, 1980
19. Steen VD, Powell DL, Medsger TA Jr: Clinical correlations and prognosis based on serum autoantibodies in
patients with systemic sclerosis. Arthritis Rheum 31:
196-203, 1988
20. Segal MR: Regression trees for censored data. Biometrics 44:3547, 1988
21. Segal MR, Bloch DA: A comparison of estimated proportional hazards models and regression trees. Stat Med
8:539-550, 1989
22. BMDP Statistical Software, revised printing. Edited by
WJ Dixon. Los Angeles, UCLA Press, 1983
23. Peters-Golden M, Wise RA, Hochberg MC, Stevens
MB, Wigley FM: Carbon monoxide diffusing capacity as
predictor of outcome in systemic sclerosis. Am J Med
77: 1027-1034, 1984
24. Rowell JR: The prognosis of systemic sclerosis. Br J
Dermatol95:57-60, 1976
25. Silman A, Jannini S, Symmons D, Bacon P: An epidemiological study of scleroderma in the West Midlands. J
Rheumatol 27:286-290, 1988
26. McCarty GA, Rice JR, Bembe ML, Barada FA Jr:
Anticentromere antibody: clinical correlations and association with favorable prognosis in patients with scleroderma variants. Arthritis Rheum 26: 1-7, 1983
27. Kostis JB, Seibold JR, Turkevich D, Masi AT, Grau
RG, Medsger TA Jr, Steen VD, Clements PJ, Szydlo L,
D’Angelo WA: Prognostic importance of cardiac arrhythmias in systemic sclerosis. Am J Med 84:10071015, 1988
28. Barnett AJ, Miller MH, Littlejohn GO: A survival study
of patients with scleroderma diagnosed over 30 years
(1953-1983): the value of a simple cutaneous classification in the early stages of the disease. J Rheumatol
15:276-282, 1988
29. Tuffanelli DL, Winkelmann RF: Diffuse systemic scleroderma: a comparison with acrosclerosis. Ann Intern
Med 57:198-203, 1962