close

Вход

Забыли?

вход по аккаунту

?

The Time-Dependence of Long-Term Prediction in Lupus Nephritis.

код для вставкиСкачать
ARTHRITIS & RHEUMATISM Volume 37
Number 3, March 1994, pp 359-368
0 1994, American College of Rheumatology
359
THE TIME-DEPENDENCE OF
LONG-TERM PREDICTION IN LUPUS NEPHRITIS
JOHN M. ESDAILE, MICHAL ABRAHAMOWICZ, TODD MAcKENZIE,
JOHN P. HAYSLETT, and MICHAEL KASHGARIAN
Objective. To assess the clinical, laboratory, and
renal biopsy predictors of long-term outcome in lupus
nephritis and to investigate the time-dependence of these
predictors.
Methods. Eighty-seven lupus nephritis patients
were studied retrospectively for the outcomes renal
failure and fatality due to renal involvement. In addition
to a conventional Cox model analysis, a new generalized
time-dependent analytic approach was developed and
used to assess the time-dependence of a predictor variable’s importance.
Results. The mean followup time was 11.9 years.
Renal failure (n = 19) was significantly predicted by
measures of renal function (abnormal serum creatinine
levels, proteinuria, duration of prior renal disease) and
immunologic activity (elevated DNA binding, hypocomplementemia, and thrombocytopenia), by overall lupus
disease activity measures (le Riche index, Lupus Activity
Criteria Count), and by the activity index, the tubuloSupported in part by grants from the Arthritis Society of
Canada (85009), the Alpha Omicron Pi Foundation, the National
Sciences and Engineering Research Council of Canada, and the NIH
(grants AR-36308, AI-07306, and AR-07530).
John M. Esdaile, MD, MPH: Department of Medicine,
McGill University, Montreal, Canada, Senior Research Scholar,
Quebec Medical Research Council, and Visiting Professor, Brigham
and Women’s Hospital, Robert B. Brigham MAMDC, Boston,
Massachusetts; Michal Abrahamowicz, PhD: Department of Epidemiology and Biostatistics, McGill University, and Senior Research
Scholar of the Quebec Medical Research Council; Todd MacKenzie, MSc: McGill University; John P. Hayslett, MD: Department of
Medicine, Yale University, New Haven, Connecticut; Michael
Kashgarian, MD: Department of Pathology, Yale University.
Address reprint requests to John M. Esdaile, MD, Montreal
General Hospital, 1650 Cedar Avenue, Montrkal, Quebec, Canada
H3G 1A4.
Submitted for publication March 17, 1993; accepted in
revised form August 12, 1993.
interstitial index, and the amount of subepithelial deposits on renal biopsy. In general, the laboratory predictors
were significantly better prognostic markers in the early
years after biopsy, the disease activity measures were
best in the later years, and the biopsy variables were
significant predictors over the entire observation period.
In contrast to the renal failure outcome, the best predictors for death not directly related to lupus nephritis
(n = 17) were the extent of comorbid diseases (principally vascular diseases), older age, and the chronicity
index. All three predicted well over the extended observation period.
Conclusion. The major predictor variables for
renal outcomes and nonrenal outcomes are distinct. The
time-dependence of the predictive ability of some variables may be important in managing individual patients.
The new generalized time-dependent analytic technique
may have widespread application in studies to identify
prognostic factors for established disease or risk factors
for the development of disease.
Over the last decade, a number of prognostic
studies of long-term outcome in lupus nephritis have
been conducted (1-19). The studies have drawn attention to the importance of markers of renal involvement
such as serum creatinine levels (2,7,8,11,13,14,16,17)
and 24-hour urinary protein excretion (6-8), as well as
other markers such as onset at a younger age (2,8,10),
the presence of hypertension and other comorbid
diseases (1,8,13,14,17), patient response to treatment
(1,6,10,13,19), and measures of systemic lupus erythematosus (SLE) disease activity (18). In addition to the
World Health Organization (WHO) classification sys-
ESDAILE ET AL
360
tem for assessing the results of renal biopsies, newer
biopsy markers of prognosis, including measures of
activity (2,7,8,11,12,14), chronicity (2,8,1&12,14), and
tubulointerstitial involvement (3,8), have been shown
to be useful in lupus nephritis.
The results of these studies have not always
been concordant. Reasons for this may include differences in patient selection, sample size, and therapy, as
well as differences in the actual outcomes studied, the
analytic techniques used, and the observation period
over which prognosis was evaluated. Differences in
observation period have not been studied. To do so we
have obtained particularly long followup information
on a large cohort of patients with lupus nephritis and
developed new flexible analytic regression techniques
to assess whether a predictor variable has impact over
the complete observation period or only a portion of it.
We have assessed clinical, laboratory, and biopsy
variables for t w o outcomes, renal failure and fatality.
PATIENTS AND METHODS
Study design and followup. All patients with SLE at
the Yale-New Haven Hospital who underwent their first
renal biopsy between January 1967 and December 1983 were
included in this retrospective cohort study. The detailed
methods for obtaining the clinical data have been published
previously (8). This earlier report evaluated followup data to
December 1984. For the present study, the observation
period on all subjects was extended to December 1990.
Three patients had been lost to followup, in 1985, 1985, and
1986, respectively. Data on mortality for these 3 were
obtained through the use of the National Death Index (20).
All 3 patients had died. In addition, patients who were lost to
followup in the earlier study (8) were tracked. The zero time
for the study was the day of the first renal biopsy.
Baseline predictor variables. Baseline predictor variables included demographic data on age, sex, race, marital
status, and socioeconomic status (21). The durations of renal
disease and of SLE were determined as described previously
(8). Information on clinical features at the date of biopsy was
obtained for lupus headache, cytoid bodies, dry eyes, dry
mouth, lymphadenopathy, hepatomegaly, splenomegaly,
myositis, cranial or motor neuropathy, affective disorder,
stroke, Raynaud’s phenomenon, and pregnancy. Vasculitis
was defined as the presence of palpable purpura, digital
ulcers, nondigital vascular ulcers, mononeuritis multiplex,
or vasculitis proven by muscle biopsy. Serositis was considered present if pleuritis, pericarditis, or peritonitis due to
SLE was present. Hypertension was defined as a blood
pressure of at least 160/95 mm Hg at the time of biopsy.
Overall SLE disease activity was assessed by the Lupus
Activity Criteria Count (22), the SLE Disease Activity Index
(23), and the National Institutes of Health (NIH) le Riche
index (8,18,24). Overall disease severity was determined by
the Ginzler index (25) and comorbidity by the Kaplan-
Feinstein index (26). Because of the small number of patients
with life-threatening (level 3) comorbid disease according to
the Kaplan-Feinstein index, data on patients in levels 2 and
3 were combined for the comorbidity analysis.
Laboratory variables assessed included hematocrit,
platelet count, erythrocyte sedimentation rate (ESR), serum
albumin, C3, C4, and, after 1973, DNA binding. Data on
blood urea nitrogen, serum creatinine, 24-hour urinary protein excretion, and urinalysis were obtained. Nephrotic
syndrome was defined as a serum albumin level <3.0 g d d l
or peripheral edema, and 24-hour urinary protein excretion
23.0 gm.
Renal biopsies were assessed by light microscopy for
WHO class (I = normal or minimal disease; I1 = mesangial
nephritis; 111 = focal proliferative nephritis; IV = diffuse
proliferative nephritis; V = membranous nephritis). The
activity and chronicity indices were determined using the
NIH version of these scales (2); the tubulointerstitial index
employed has been described previously (8). Ultrastructural
study data were missing for 3 patients (8). The original tissue
blocks on these patients were located and processed so that
ultrastructural studies were complete. They were used to
classify the presence of electron-dense deposits (levels 0,
trace, 1, 2, 3) in the mesangium and the subepithelial and
subendothelial glomerular basement membrane regions (8).
Outcomes. The outcomes studied were renal failure
(end-stage renal disease requiring dialysis or transplantation)
or death due to SLE. The deaths were classified into those
attributable to renal disease and those not attributable to
renal disease (8,18). For brevity, we present detailed results
only for the outcomes, renal failure and death not due to
renal involvement.
Analysis. The ability of each of the clinical variables
to predict each outcome was investigated using a univariate
Cox proportional hazards model (27). This model expresses
the strength of the association between the predictor and the
outcome in terms of a hazard ratio. The hazard ratio may be
conceptualized as a risk ratio. The Cox model assumes that
the prognostic impact of a predictor is constant over the
entire observation period of the study (27). For example, if
3% of lupus nephritis patients and 1% of SLE patients
without nephritis are estimated to die in the first year of
observation, the estimated hazard ratio equals 3.0 and the
same ratio is assumed to characterize the relative risk
(nephritis present versus absent) throughout the entire followup period. Thus, if, for example, 2% of patients without
nephritis who survived until the tenth year are estimated to
die during this year, the corresponding risk of dying among
the survivors in the lupus nephritis group is “automatically”
set at 6% (2 x 3).
Often the assumption of a constant hazard ratio
(constant prognostic ability) is clinically plausible, but a
priori it need not be universally true. In some situations, the
strength of the association may become gradually weaker
over time, indicating a decreasing prognostic ability of a
variable. In other cases, a predictor may have long-term
predictive ability but may not be useful for predicting
outcomes in the early phase of the followup period. In these
situations, the conventional Cox model will not be able to
account for the relevant features of the data and may lead to
invalid conclusions. To address this issue, we have devel-
36 1
LONG-TERM PREDICTORS IN LUPUS NEPHRITIS
oped a generalization of the Cox model in which the predictive ability of a variable is allowed to change over time.
Accordingly, the hazard ratio becomes a function of time
rather than a single constant. We have employed regression
splines (28-3 1) to estimate this function.
A number of generalized time-dependent regression
spline models of different complexities were considered and
the model most consistent with empirical data was selected
based on the Akaike Information Criterion (AIC) (32). This
criterion rewards the model’s goodness-of-fit to empirical
data while penalizing for increasing model complexity.
Therefore, it favors the simplest models, all other things
being equal. We used the standard likelihood ratio test to
determine whether the AIC-optimal model fits the actual
data significantly better than the conventional Cox model. A
significant result has been interpreted as evidence of the
time-dependence of the prognostic ability of a predictor. In
the case of significant results, we have plotted the estimated
hazard ratio as a function of the duration of followup to
establish how the variable’s impact changes over time. The
95% confidence intervals for the estimated function were
then used to assess the significance of the association at any
specific time.
This approach allowed us to determine whether a
variable was significant over the entire period of observation
or was a useful predictor over only a limited time interval. In
the latter case, we classified the relevant variable as early
(0-5 years), middle ( 6 1 0 years), or late (11-15 years). This
classification was selected to balance the number of outcome
events in each period, i.e., number of renal failure events:
early = 7 , middle = 6, late = 6; number of nonrenal deaths:
early = 4, middle = 8, late = 5. Although the number of
events is larger than in many studies on this topic, the
numbers are small in absolute terms. Therefore, our analyses should be considered exploratory. The results need to be
confirmed with an independent data set.
RESULTS
Clinical and biopsy characteristics. Eighty-seven
subjects met the inclusion criteria. Data were available
up to the time of death or to December 1990 for 12,372
patient-months, out of a potential observation period
of 12,441 patient-months (99.5%). Thus, the average
followup to death or 1990 was 11.9 years (range
0.3-23.0 years).
Of the 87 patients, 75 (86%) were female, 73
(84%) were white, 49 (56%) had never been married,
and 53 (61%) were in the top 3 of the 6 socioeconomic
categories. The average age was 27 years, and 18 of
the patients (21%) were children (515 years). SLE was
diagnosed a median of 6 months before and renal
disease was detected a median of 2 months before
renal biopsy. At the time of renal biopsy, 10 patients
(11%) had a blood pressure of at least 160/95 mm Hg;
Raynaud’s phenomenon was present in 20 (23%),
vasculitis in 10 (ll%), and serositis in 6 (7%). Forty-
Table 1. Characteristics of patients with lupus nephritis at the time
of renal biopsy
~~
Variable*
Laboratory characteristics
Serum creatinine (mgldl)
24-hour urinary protein (gm)
Platelet count ( x 1,000/mm3)
ESR (mdhour)
DNA binding (%)
Serum C3 (rng/dl)
Indices
le Riche disease activity
index
Lupus Activity Criteria
Count
Ginzler severity index
Kaplan-Feinstein comorbidity
index
Biopsy characteristics
NIH activity index
NIH chronicity index
Tubulointerstitial index
No. (%) No. (%)
below
above or
Cutpoint cutpoint at cutpoint
1.5
3.0
140
50
75
61
66 (76)
61 (74)
6 (7)
28 (48)
41 (71)
57 (66)
21 (24)
22 (26)
75 (93)
31 (52)
17 (29)
30 (34)
4
60 (72)
23 (28)
3
41 (48)
45 (52)
7
1
43 (49)
44 (51)
44 (51)
43 (49)
9
3
1
59 (68)
57 (66)
28 (32)
30 (34)
* ESR = erythrocyte sedimentation rate; NIH = National Institutes
of Health.
three (49%) had at least one comorbid condition; these
were principally vascular (36 of the 43 [84%1 had
hypertension, peripheral vascular disease, or coronary
disease). Seventy-six percent of the patients had anemia and 24% had leukopenia. Other selected clinical
and laboratory measures are described in Table 1.
Tissue for light microscopic and ultrastructural
studies was available from all 87 patients. WHO class
I disease was present in 1patient, class I1 (mesangial)
in 13 (15%), class 111 (focal proliferative) in 5 (6%),
class IV (diffuse proliferative) in 59 (68%), and class V
(membranous) in 9 (10%). The NIH biopsy activity
index was low (15) in 46 patients (53%) and chronic
changes were absent in 25 (29%). Fifty-seven (66%)
had an abnormal tubulointerstitial index. Electrondense deposits were absent (level 0) or present in a
marked amount (levels 2 or 3) at the mesangial region
in 7 (8%) and 56 (64%) of the patients, respectively, at
the subepithelial region in 34 (39%) and 23 (26%), and
at the subendothelial region in 21 (24%) and 36 (41%).
Therapy. Therapy received by the 87 patients
has been described (8,18,33). Treatment recommendations were based on a standard protocol. Eighty-five
patients (98%) were treated with prednisone and 68
(78%) with immunosuppressive agents (azathioprine
was the initial agent used for 60 of 68). Immunosuppressive drugs were recommended particularly for
ESDAILE ET AL
362
Table 2.
Predictors of renal failure in lupus nephritis*
Time-dependent
(spline) model
Conventional Cox model
Variable, scoring
~~~
~
95%
Relative
hazard per confidence
unit change
interval
P
Specific
timing of
importance?
PS
~
Clinical and laboratory characteristics
Nephrotic syndrome, presenuabsent
Anti-DNA (%), ?75/<75
24-hour urinary protein (gm), ?3.0/<3.0
Platelet count (x 1,000/mrn3), <140/?140
Serum creatinine (mgldl), ? I S / < 1.5
Comorbidity index, 0-2
le Riche index, 0-10
C3 (mgldl), ~ 6 0 / > 6 0
Duration of prior renal disease (months),
?22/<22
Lupus Activity Criteria count, 0-7
Biopsy characteristics
Activity index, 0-24
Tubulointerstitial index, 0-2
Subepithelial deposits, 0-3
4.10
4.10
3.13
4.14
3.08
2.14
1.32
2.97
1.95
1.57-10.66
1.42-11.8
1.27-7.72
1.20-14.22
1.15-8.30
1.104.18
1.02-1.70
0.98-8.97
0.55-6.88
0.002
0.005
0.009
0.02
0.02
0.02
0.03
0.04
NSB
1.31
0.81-2.13
NS
1.13
2.01
1.77
1.03-1.24
1.12-3.61
1.08-2.88
0.006
0.02
0.02
-
-
-
Early
0.04
Early
Late
0.03
0.03
Early
0.007
Late
0.02
-
-
-
-
-
-
* Renal failure occurred in 19 of the 87 lupus nephritis patients studied. Comorbidity index level 3 has
been combined with level 2.
t A dash indicates that no time-dependent model was superior to the conventional Cox model.
$ The likelihood ratio P value corresponding to the test that the best-fitting spline model provides
additional information to that provided by the conventional Cox model.
5 NS = not significant.
Cox and generalized time-dependentmodels. Tables 2 and 3 provide summaries of the results for the
outcomes renal failure (19 events) and death not
attributable to renal involvement (17 events). For
brevity, the results for deaths attributable to renal
involvement (n = 17) and for overall mortality due to
patients with marked subendothelial deposits. These
agents were continued for at least 3 years in 85% of the
68 patients. Thirty-four (94%) of 36 patients with level
2 or 3 subendothelial deposits received these agents,
versus 34 (67%) of 51 patients with lesser grades for
this variable (P< 0.005).
Table 3. Predictors of death not due to renal involvement in lupus nephritis*
Time-dependent
(spline) model
Conventional Cox model
Variable, scoring
Clinical and laboratory characteristics
Comorbidity index, 0-2
Age (years), r40/<40
Raynaud’s phenomenon, presentlabsent
ESR (mmhour), ?50/<50
Socioeconomic status, 1-6
c3, s60/>60
Serositis, presenuabsent
Biopsy characteristics
Chronicity index, ?31<3
SubeDithelial deDosits. 0-3
Relative
95%
hazard per confidence
unitchange
interval
P
Specific
timing of
importancet
PS
3.69
6.89
3.05
4.81
1.48
1.10
2.27
1.38-7.26
2.58-18.40
1.767.90
1.02-27.70
1.00-2.19
0.41-2.94
0.63-8.15
<0.001
<0.001
0.02
0.03
0.05
NS
NS
Late
-
0.02
-
Early
Late
0.007
0.02
2.72
1.56
1.03-7.20
0.92-2.64
0.03
0.09
Early
0.01
-
-
* Death not due to renal involvement occurred in 17 of the 87 lupus nephritis patients studied.
Comorbidity index level 3 has been combined with level 2. ESR = erythrocyte sedimentation rate; NS
= not significant.
t A dash indicates that no time-dependent model was superior to the conventional Cox model.
S The likelihood ratio P value corresponding to the test that the best-fitting spline model provides
additional information to that provided by the conventional Cox model.
363
LONG-TERM PREDICTORS IN LUPUS NEPHRITIS
SLE are not presented. The results for the former
were almost identical to those for renal failure; for the
latter, results were similar to those with the combination of predictors for renal and nonrenal deaths, as
would be expected. The percentage of patients without
renal failure at 5, 10, and 20 years was 91%, 83%, and
70%, respectively. The percentage surviving at 5 , 10,
and 20 years was 84%, 72%, and 52%, respectively.
For the 17 deaths not due to renal involvement, the
underlying cause was vascular (e.g., myocardial infarction, cerebrovascular accident) in 12, vasculitis in
3, and infection in 2.
Tables 2 and 3 show the results derived from
the conventional Cox model and the generalized timedependent model, for all of the clinical, laboratory,
and biopsy variables described in Patients and Methods for which a statistically significant result was
obtained. Variables not included in the tables were not
significant with either analytic approach. The conventional Cox model data are the results for predictor
variables under the proportional hazards assumption
required by this model, i.e., the predictor is assumed
to have a constant effect throughout the entire followup period after renal biopsy (almost 12 years on
average). The relative hazard is per unit change in the
variable and is akin to a relative risk. The timedependent model data are the results for those variables that were significantly better predictors for a
portion of the observation period than for the complete
observation period. The time period during which the
variable was a significant predictor is reported, as is
the P value for the test of significance that the generalized time-dependent model is better than the conventional Cox model.
In several cases, the P value reported in the last
column of Table 2 represents a significant result rejecting the conventional Cox model in favor of models
with time-dependent relative hazards. This provides
evidence that the prognostic ability of a given variable
changes during the observation period. In such cases,
the conventional Cox model analyses are invalid. For
example, in the case of the duration of renal disease
prior to study onset, defined as a binary variable (<22
months versus 2 2 2 months), the test for timedependence gives a significant result (P = 0.007) and
therefore the constant hazard ratio of 1.95 obtained
from the Cox model is potentially misleading.
To contrast the Cox and the time-dependent
models, the results of both analyses for the variable
duration of prior renal disease are plotted in Figure 1.
The horizontal axis corresponds to the observation
L
1
O.'
°
7
i
0
5
10
15
10
15
YEARS
B
100
1
0
5
YEARS
Figure 1. The effect of long duration of renal disease (222 months)
on the risk of renal failure, estimated using A, the conventional Cox
model (not significant) and B, the time-dependent (spline) model.
The spline model fits significantly better than the conventional Cox
model (P = 0.007). The vertical axis represents the hazard ratio
(relative risk) and the horizontal axis represents time in years after
the first renal biopsy. The thick solid line is the hazard ratio, the thin
solid lines are the 95% confidence intervals for the hazard ratio, and
the dashed line denotes a hazard ratio of unity.
period, and along the vertical axis the hazard ratio is
plotted on a logarithmic scale. Figure 1A presents the
Cox model estimates: the thick line corresponds to the
constant hazard ratio of 1.95 and the 2 boundaries
define the 95% confidence interval (from 0.55 to 6.88).
Because the confidence interval includes 1 .O, duration
of renal disease does not appear to be a significant
predictor of renal failure in the conventional Cox
model (the ratio of 1.0 means that the risks are the
same in both groups).
In contrast, in the spline model (Figure lB),
both the estimated hazard ratio and its confidence
ESDAILE ET AL
364
intervals vary as a function of time. The estimated
function indicates that the relative risks associated
with longer duration of disease are very high in the first
2 years after biopsy, and that this risk increase is
statistically significant because the lower boundary of
the confidence interval over this period is close to 1.8.
Thus, the spline model indicates that there is reliable
evidence that during the first 2 years of followup, the
risk of renal failure among patients with longer duration of disease is at least 80% higher than for those
with shorter disease duration. However, Figure 1B
also shows that the hazard ratio quickly decreases
with increasing time of observation and by the fourth
year it becomes nonsignificant (since the confidence
intervals include 1.0). Accordingly, there is no evidence that duration of prior disease reliably predicts
renal failure 4 years or more after biopsy.
To summarize, in contrast to the nonsignificant
finding of the Cox model, the time-dependent model
established duration of renal disease as a significant
and very powerful predictor of early renal failure while
suggesting that it has no association with risk of later
renal failure. This suggests that, based on the conventional Cox analysis, a study with a relatively short
followup (e.g., 5 years) would likely find the duration
of prior renal disease to be highly significant. In
contrast, in a study with a longer followup, such as
ours, the inability of the conventional model to account for the decreasing prognostic ability of renal
disease duration would result in a potentially misleading conclusion.
Regression splines based on time-dependent
models can also detect that a predictor found to be
significant in the conventional Cox model in fact
changes its impact over time. This situation is illustrated in Figure 2, where elevated 24-hour urinary
protein excretion is evaluated as a predictor of renal
failure. According to the Cox model, patients with
abnormal values of this variable at the time of biopsy
are at -3 times higher risk throughout the 15-year
followup period (Figure 2A and Table 2). The fact that
the 95% confidence limits in Figure 2A do not include
1.0 suggests that the association is significant during
the entire period. However, the likelihood ratio test
rejected the hypothesis of a constant hazard ratio in
favor of a time-dependent association (P= 0.04; Table
2). This indicates that the spline model selected by the
AIC criterion is more consistent with the data. The
hazard ratio estimated by this model is plotted in
Figure 2B. The estimated hazard ratio during the first
few years after biopsy is much higher than in the
-~
Y
0014-
,
,
,
I
,
7
1
.
I
5
0
,
. .
I
15
10
YEARS
0.014
0
.
,
I
I
,
I
I
I
.
,
I
. .
,
10
5
YEARS
Figure 2. The effect of abnormal urinary protein excretion (23.0
gdday) on the risk of renal failure, estimated using A, the conventional Cox model (P = 0.009), and B, the spline model. The spline
model fits significantly better (P = 0.04). See Figure 1 for details.
conventional Cox model, suggesting that the latter
underestimates the prognostic utility of abnormal urinary protein excretion in the early period. In contrast,
the Cox model tends to overestimate the impact of this
variable in the later period when, according to the
spline model, the hazard ratio gradually decreases to
0, losing significance -6 years after biopsy (Figure 2B).
Similarly, the spline modeling analysis can detect predictors that are not significant in the Cox model
because their major impact on prognosis occurs well
after the renal biopsy. An example of this was the
Lupus Activity Criteria Count, which measures the
overall activity of SLE (Table 2 and Figure 3). The le
Riche index, another general disease activity measure
for SLE, was significant with both the conventional
365
LONG-TERM PREDICTORS IN LUPUS NEPHRITIS
Cox and the generalized time-dependent models (Table 2), but the results of the generalized timedependent model indicated that this index predicted
the late outcomes best.
Finally, some variables predict throughout the
observation period and there is no evidence of changes
in their predictive ability over time. The level of the
comorbidity index predicted fatality not attributable to
renal disease significantly in the Cox model (Table 3),
and none of the time-dependent models was superior
to the Cox model.
Predictors of renal and nonrenal outcomes.
Among the most powerful predictors for renal failure
were markers of renal disease severity (nephrotic
syndrome, markedly elevated protein excretion, abnormal serum creatinine levels, duration of prior renal
disease), altered immunologic activity (markedly elevated DNA binding, diminished C3 levels, and thrombocytopenia), and global SLE disease activity (Lupus
Activity Criteria Count, NIH le Riche Index) (Table
2). The (biopsy) activity index, the extent of tubulointerstitial damage, and the extent of subepithelial
deposits were significant predictors.
In contrast to the results for the outcome renal
failure, the strongest predictors for SLE deaths not
directly caused by renal involvement were the presence and severity of comorbid diseases (Figure 4) and
older age (Table 3). Comorbidity and older age were
significantly correlated with one another (r = 0.37 by
Spearman rank correlation, P = 0.005). Both were
PERCENTAGE
LEVEL 0
WITHOUT
NON-RENAL
LEVELS 1 8 2
5040-
302010O
i
,
,
,
,
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
l
strong predictors throughout the followup period. The
presence of Raynaud’s phenomenon, elevated ESR,
and low socioeconomic status predicted nonrenal
deaths, and diminished C3 and the presence of serositis also predicted this outcome for at least a portion of
the observation period. The best renal biopsy predictor was a moderately elevated chronicity index, which
predicted over the entire observation period (Table 3).
Subepithelial electron-dense deposits were an early
predictor.
DISCUSSION
Y
01
0
5
10
15
YEARS
Figure 3. The effect of a 1-unit increase in the Lupus Activity
Criteria Count on the risk of renal failure, estimated using the
time-dependent spline model. This model fit significantly better than
the conventional Cox model (P = 0.02). See Figure 1 for details.
The predictors of renal failure and of death not
attributable to renal disease have been described for a
cohort of patients with lupus nephritis for whom there
was a long period of followup (average 11.9 years,
range 0.3-23.0 years). The strongest predictors for the
renal failure outcome were variables that reflected the
severity of renal involvement, the degree of immunologic activity, and global SLE disease activity. At least
for renal failure, many of the renal and immunologic
markers predicted the outcome in the first 5 years of
observation (Figures 1 and 2), whereas measures of
disease activity indices predicted the development of
renal failure after 10 years or longer (Figure 3). We
have noted previously that global SLE disease activity
is an important predictor among lupus nephritis patients who lack the markers for severe renal disease
(18). The present findings show the distinct timedependence of the prognostic import of renal involvement and global disease activity in patients with lupus
366
nephritis. Patients with severe renal disease develop
renal failure early, whereas those with very active
SLE overall are at risk later.
The activity index was an important biopsy
predictor of renal failure. This is consistent with the
results of previous studies (2,7,8,11,12,14). The tubulointerstitial index was also a significant predictor for
this outcome. Tubulointerstitial disease, particularly
damage, increasingly has been recognized as an important determinant of renal outcome in recent years
(3,8,34-37).
No studies have evaluated lupus nephritis
deaths not due to renal involvement, other than as a
component of total mortality. The only renal or immunologic marker discussed above that was a significant
predictor for SLE deaths not attributable to renal involvement was hypocomplementemia. The most powerful predictors were the presence of comorbid disease and
older age (Table 3). The effect of age likely reflects the
higher frequency of comorbid diseases with aging in that
the 2 variables were correlated. Comorbid disease, particularly hypertension (1,8,13,14)and vascular diseases,
perhaps related to treatment with prednisone (38),
have been recognized as important prognostic markers. The significant association of Raynaud’s phenomenon is of interest given this increased interest in
vascular involvement in SLE.
The principal biopsy predictor of nonrenal
deaths also differed from the biopsy predictors associated with renal outcomes. Renal outcomes were
predicted by the activity index, the tubulointerstitial
index, and subepithelial deposits. For nonrenal
deaths, the chronicity index was the only biopsy
predictor that was reliable for the complete observation period. Previously, the chronicity index had been
associated with both renal failure (2,10,11,14) and
mortality (10-12) in some studies, but not all
(6,7,9,13).
Subendothelial deposits were an important indicator of the severity of glomerular diseases in our
previous studies (39). In the current study, subepithelial deposits were a strong predictor of both outcomes.
The presence of subepithelial deposits is consistent
with the deposition of immune complexes involving
cationic antigens such as Sm, Ro, and La. A recent
report by Tokano and coworkers (40) demonstrated
that antibodies to these antigens are more likely to be
of IgGl or IgG3 class. IgGl and IgG3 bind complement
and hence are more likely to be associated with severe
scarring forms of lupus glomerulonephritis. Therefore,
the presence of subepithelial deposits in patients with
ESDAILE ET AL
WHO class IV lupus nephritis and high activity indices
suggests that not only is the immunologic response
active, but also that it is more pathogenic to glomeruli,
and thus predictive of poor renal outcome.
The distinction in the results for renal outcomes
and nonrenal outcomes has important implications for
the management of lupus nephritis. Prompt therapy
and control of the lupus diathesis may be expected to
reduce the frequency of renal failure. However, reduction in the frequency of nonrenal outcomes may require
greater attention to other factors such as smoking
cessation and the control of hypertension, hypercholesterolemia, and major obesity (38). Prednisone
therapy is associated with hypertension and myocardial infarction in SLE and is a mainstay in the treatment of lupus nephritis. This suggests that treatment
with immunosuppressive agents may be beneficial in
severe lupus nephritis both by improving renal outcome and by reducing the requirement for prednisone
and, thereby, the potential for nonrenal adverse
events.
A second aspect of this study was the evaluation of the time-dependent nature of the prognostic
importance of predictor variables. Some of the variables predicted well throughout the long observation
period of this study. Others predicted better for a
portion of the observation period, although they remained significant predictors for the complete observation period. Most importantly, some variables were
powerful predictors for only a portion of the followup
period, and their prognostic ability would not have
been detected at all by the conventional Cox model.
This type of analysis has not been employed previously in studies of prognosis in lupus nephritis or other
diseases.
The specific results of the current analysis have
not been validated. Validation would require confirmation in a second data set with a similar number of
outcome events. Nonetheless, the general findings are
of potential importance. Although the two most powerful predictors of nonrenal death, i.e., comorbidity
and older age, predict throughout the observation
period (i.e., their prognostic ability does not change
over time), many of the clinical and laboratory predictors of renal outcome are most powerful in the early
years of study (generally the first 5 years). Depending
on the duration of the study, the sample size, and the
frequency and timing of outcome events, the importance of these predictors could be missed. The use of
analytic techniques that assess the timing of the prognostic impact of a marker variable might be expected
367
LONG-TERM PREDICTORS IN LUPUS NEPHRITIS
to reduce some of the conflict among study results that
is apparent in the published literature. Furthermore,
the time-dependent model detected the importance of
overall measures of SLE disease activity for predicting
late renal failure and the late effect of Raynaud’s
phenomenon on nonrenal deaths, effects that would
have gone undetected if flexible modeling techniques
had not been used.
This new analytic technique may be helpful in
the study of prognosis of many rheumatic diseases
(e.g., predictors of mortality in rheumatoid arthritis).
It would also be valuable in studying risk factors for
the development of rheumatic diseases in long-term
cohort studies (e.g., the risk factors for the development of osteoarthritis). In studies of both prognosis
and disease development, it has the potential to provide important insights into associations that are frequently complex.
In summary, we have evaluated the predictors
for very long-term outcome in lupus nephritis. The
particularly long followup after the initiation of the
study permitted evaluation of the time-dependent or
time-independent impact of predictor variables. The
relevance of renal, immunologic, and global disease
activity markers for renal outcome and the importance
of comorbid diseases for nonrenal events is noted.
This suggests that future improvement in the prognosis
for lupus nephritis will require treatment protocols
designed to alter factors that affect both renal and
nonrenal manifestations of this disease.
REFERENCES
1. Wallace DJ, Podell TE, Weiner JM, Cox MB, Klinenberg JR,
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
ACKNOWLEDGMENTS
Many individuals assisted in obtaining the nearly
complete followup data for this study. Norman J. Siegal,
MD, and the many patients who provided data or agreed to
be reevaluated by their physicians deserve special mention.
Robert Bilgrad of the National Death Index was of considerable help. The following physicians kindly assisted in
providing followup data: Drs. Walter Anyan, Stephen Atlas,
David A. Baggish, Daniel C. Belin, Margaret Bia, Murray
Brodoff, Dorothy Estes, Luis Fernandez-Herlihy, Richard
Fearon, Fredric 0. Finklestein, Geoffrey Gladstein, Robert
Gordon, Robert Gray, Moreson Kaplan, James Kenney,
Alan S. Kliger, Daniel J. McCarty, Susan McIntosh, Lewis
Parker, T. G. Patel, Russell Randall, Christopher Reid,
David Satloff, Robert T. Schoen, Michael Shore, Ann Skopek, Alan M. Spagnola, Ralph Tella, Edward H. Scherr,
Mark Schwartz, David Trostle, Joanne Wilkinson, and Kimberly Wilson. Ms. Barbara Cont provided expert secretarial
assistance throughout all phases of the study, including
preparation of the manuscript.
12.
13.
14.
15.
16.
17.
Forouzesh S, Dubois EL: Lupus nephritis: experience with 230
patients in a private practice from 1950 to 1980. Am J Med
72:209-220, 1982
Austin 111HA, Muenz LR, Joyce KM, Antonovych TA, Kullick
ME, Klippel JH, Decker JL, Balow JE: Prognostic factors in
lupus nephritis: contribution of renal histologic data. Am J Med
75~382-391, 1983
Austin I11 HA, Muenz LR, Joyce KM, Antonovych TT, Balow
JE: Diffuse proliferative nephritis: identification of specific
pathologic features affecting renal outcome. Kidney Int 29689695, 1984
Banfi G, Mazzucco G, Di Belgiojoso GB, Bosisio MB, Stratta P,
Confalonieri R, Ferrario F, Imbasciati E, Monga G: Morphological parameters in lupus nephritis: their relevance for classification and relationship with clinical and histological findings
and outcome, Q J Med 55:153-168, 1985
Stamenkovic I, Favre H, Donath A, Assimacopoulos A, Chatelanat F: Renal biopsy in SLE irrespective of clinical findings:
long-term follow-up. Clin Nephrol26: 109-1 15, 1986
Appel GB, Cohen DJ, Pirani CL, Meltzer JI, Estes D: Longterm follow-up of patients with lupus nephritis: a study based on
the classification of the World Health Organization. Am J Med
83:877-885, 1987
Magill AB, Puterman ML, Ballon HS, Chan V, Lirenman DS,
Rae A, Sutton RAL: Prognostic factors in diffuse proliferative
lupus glomerulonephritis. Kidney Int 3451 1-517, 1988
Esdaile JM, Levinton C, Federgreen W, Hayslett JP, Kashgarian M: The clinical and renal biopsy predictors of long-term
outcome in lupus nephritis: a study of patients and review of the
literature. Q J Med 72:779-833, 1989
Schwartz MM, Bemstein J, Hill GS, Holley K, Phillips EA, the
Lupus Nephritis Collaborative Study Group: Predictive value of
renal pathology in diffuse proliferative lupus glomerulonephritis. Kidney Int 39S91-896, 1989
Laitman RS, Glicklich D, Sablay LB, Grayzel AI, Barland P,
Bank N: Effect of long-term normalization of serum complement levels on the course of lupus nephritis. Am J Med
87:132-138, 1989
Nossent HC, Henzen-Logmans SC, Vroom TM, Berden JHM,
Swaak TJG: Contribution of renal biopsy data in predicting
outcome in lupus nephritis: analysis of 116 patients. Arthritis
Rheum 33:970-977, 1990
McLaughlin J, Gladman DD, Urowitz MB, Bombardier C,
Farewell VT, Cole E: Kidney biopsy in systemic lupus erythematosus. 11. Survival analyses according to biopsy results.
Arthritis Rheum 34:1268-1273, 1991
Levey AS, Lan S-P, Corwin HD, Kasinath BS, Lachin J,
Neilson EG, Hunsicker LG, Lewis EJ, and the Lupus Nephritis
Collaborative Study Group: Progression and remission of renal
disease in the lupus nephritis collaborative study: results of
treatment with prednisone and short-term oral cyclophosphamide. Ann Intern Med 116:114-123, 1992
McCurdy DK, Lehman TJA, Bemstein B, Hanson V, King KK,
Nadorra R, Landing BH: Lupus nephritis: prognostic factors in
children. Pediatrics 89:240-246, 1992
Esdaile JM, MacKenzie T, Barrk P, Danoff D, Osterland CK,
Somerville P, Quintal H, Kashgarian M, Suissa S: Can experienced clinicians predict the outcome of lupus nephritis? Lupus
1~205-214, 1992
Derksen RHWM, Hen6 RJ, Kater L: The long-term clinical
outcome of 56 patients with biopsy-proven lupus nephritis
followed at a single center. Lupus 1:97-103, 1992
Gruppo Italian0 per lo Studio della Nefrite Lupica (GISNEL):
Lupus nephritis: prognostic factors and probability of maintain-
368
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
ing life-supporting renal function 10 years after the diagnosis.
Am J Kidney Dis 19:473479, 1992
Goulet JR, MacKenzie T, Levinton C, Hayslett JP, Ciampi A,
Esdaile JM: The long-term prognosis of lupus nephritis: the
impact of disease activity. J Rheumatol 20:59-65, 1993
Fraenkel L, MacKenzie T, Joseph L, Kashgarian M, Hayslett
JP, Esdaile JM: Response to therapy as a predictor of outcome
in lupus nephritis (abstract). Arthritis Rheum 35 (suppl9):S180,
1992
McMahon B: The national death index. Am J Public Health
73547-648, 1983
Wilkinson RG (ed). Class and Health: Research and Longitudinal Data. New York, Tavistock Publications, 1986
Urowitz MB, Gladman DD, Tozman ECS, Goldsmith CH: The
lupus activity criteria count. J Rheumatol 11:783-787, 1984
Bombardier C, Gladman DD, Urowitz MB, Caron D, Chang
CH, the Committee on Prognosis Studies in SLE: Derivation of
the SLEDAI: A disease activity index for lupus patients.
Arthritis Rheum 35:630-640, 1992
Le Riche NGH, Muenz LR, Silverman JA, Decker JL: Development of an SLE activity index. Submitted for publication
Ginzler EM, Diamond HS, Weiner M, Schlesinger M, Fries JF,
Wasner C, Medsger TA Jr, Ziegler G, Klippel JH, Hadler NM,
Albert DA, Hess EV, Spencer-Green G, Grayzel A, Worth D,
Hahn BH, Barnett EV: A multicenter study of outcome in
systemic lupus erythematosus. I. Entry variables as predictors
of prognosis. Arthritis Rheum 25:601411, 1982
Kaplan MH, Feinstein AR: The importance of classifying initial
co-morbidity in evaluating the outcome of diabetes mellitus. J
Chronic Dis 27:387404, 1974
Cox DR: Regression models and life tables (with discussion). J
R Stat SOCB34:187-220, 1972
Ramsay J: Monotone splines in action (with discussion). Stat
Sci 3:425461, 1988
Ramsay JO, Abrahamowicz M: Binomial regression with monotone splines: a psychometric application. J Am Stat Assoc
84:90&915, 1989
ESDAILE ET AL
30. Sleeper LA, Hamngton D P Regression splines in the Cox
model with application to covariate effects in liver disease. J Am
Stat Assoc 85:941-949, 1990
31. Abrahamowicz M, Ciampi A, Ramsay JO: Non-parametric
density estimation for censored survival data: regression splines
approach. Can J Stat 20:171-185, 1992
32. Akaike H: A new look at the statistical model identification.
IEEE Trans Automatic Control AC19:71&723, 1974
33. Esdaile JM, Federgreen W, Quintal H, Suissa S, Hayslett JP,
Kashgarian M: Predictors of one year outcome in lupus nephritis: the importance of renal biopsy. Q J Med 8197-918, 1991
34. Schwartz MM, Fennel1 JS, Lewis EJ: Pathologic changes in the
renal tubule in systemic lupus erythematosus. Hum Pathol
13534547, 1982
35. Park MH, Agati VD, Appel GB, Pirani CL: Tubulointerstitial
disease in lupus nephritis: relationship to immune deposits,
interstitial inflammation, glomerular changes, renal function,
and prognosis. Nephron 44:30%319, 1986
36. Alexopoulos E, Seron D, Hartley RB, Cameron JS: Lupus
nephritis: correlation of interstitial cells with glomerular function. Kidney Int 37:lOO-109, 1990
37. Nath KA: Tubulointerstitial changes as a major determinant in
the progression of renal damage. Kidney Int 2O:l-17, 1992
38. Petri M, Spence D, Bone LR, Hochberg MC: Coronary artery
disease risk factors in the Johns Hopkins lupus cohort: prevalence, recognition by patients and preventative practices. Medicine (Baltimore) 71:291-302, 1992
39. Esdaile JM, Joseph L, MacKenzie T, Kashgarian M, Hayslett
JP: The pathogenesis and prognosis of lupus nephritis: information from repeat renal biopsy. Semin Arthritis Rheum 23:135148, 1993
40. Tokano Y, Yasuma M, Harada S, Takasaki Y, Hashimoto H,
Okuria K, Hirose S: Clinical significance of anti-Sm and U,
RNP antibodies in patients with systemic lupus erythematosus
and mixed connective tissue disease. J Clin Immunol 11:317325, 1991
Документ
Категория
Без категории
Просмотров
0
Размер файла
952 Кб
Теги
lupus, times, terms, long, nephritis, dependence, prediction
1/--страниц
Пожаловаться на содержимое документа