PATIENTS AND METHODS
- Top of page
- PATIENTS AND METHODS
Eligibility criteria. We performed a retrospective cohort study of consecutive patients with definite primary SS. Patients with primary SS were included in the study if they had been seen at least once at the University of Ioannina or at the University of Athens between January 1981 and November 1999. Only patients who met the European Study Group criteria (8) were considered. Because the criteria were developed in 1993 and further validated in 1996 (9), patients seen before 1993 were considered retrospectively to have been diagnosed as having definite primary SS when they had sufficient evidence to satisfy the European Study Group criteria.
Followup began at the first study visit, which was either the first appointment in our clinics or the time at which a diagnosis of definite primary SS was made, whichever occurred later. Patients with probable primary SS (8, 9) were excluded from the analysis unless they eventually developed definite primary SS, in which case, followup started with the diagnosis of definite primary SS.
Both incident and prevalent cases qualified for study inclusion. For incident cases, the diagnosis of primary SS was made during the first study visit. For prevalent cases, the diagnosis of primary SS had been established elsewhere, prior to the first study visit. However, patients referred to us who had been preliminarily diagnosed elsewhere within 4 months of their first study visit were considered incident cases. Statistical modeling was performed on the complete study cohort, but analyses limited to the incident cases are also presented.
Data collection. Detailed standardized medical records have been available for all patients since 1981. Followup was censored on November 1, 1999. For patients last seen in the clinic between July 1, 1999 and November 1, 1999, followup was censored at the last clinic visit. For patients last seen before July 1, 1999, we tried to update their status. Communication was attempted using contact information available from the medical records. For the few patients on whom we could obtain no direct or indirect information about their status, we searched the death certificates of their residence area for the periods-at-risk, when a specific residence area was recorded.
The following individual patient data were collected: incident or prevalent case, sex, date of first study visit, age at first study visit, date of last followup, survival status, documentation of lymphoma or other LPD prior to death or at last known followup, date of diagnosis and type of LPD, and causes of death. In addition, we recorded information on the presence of ocular symptoms (dry eyes), oral symptoms (dry mouth), ocular signs (positive findings on rose bengal staining and/or Schirmer's test), parotid enlargement (bilateral or unilateral), palpable purpura, lymphadenopathy (excluding minor inguinal lymphadenopathy), splenomegaly, and results of laboratory tests obtained routinely (anti-Ro and anti-La antibodies, as measured by counterimmunoelectrophoresis, antinuclear antibodies, as measured by immunofluorescence, rheumatoid factor, as measured by latex fixation, and complement C3 and C4 levels, as measured by nephelometry) within the first 6 months of followup. We also recorded measurements of cryoglobulins obtained at any time during followup, at the discretion of the physician.
Statistical analyses. Time-to-event analyses for death and LPD are presented with Kaplan-Meier curves (10). The standardized mortality ratio was estimated using for comparison the age- and sex-matched probabilities of mortality from life-tables of the general population of Greece for 1990. Demographic, clinical, and laboratory characteristics routinely recorded in the initial evaluation of patients with primary SS were assessed as predictors of death and LPD by use of proportional hazards models. LPD analyses excluded the patients who already had an LPD diagnosis at the beginning of the study followup. Both univariate and multivariate models were considered. Multivariate models were built with backward elimination of variables using likelihood ratio criteria (11). Predictive models for mortality were stratified for age (<50 years, 50–59 years, 60–69 years, and >69 years). Analyses adjusting for age as a covariate yielded similar results (data not shown).
Unless specified otherwise, patients with missing data were excluded from the respective analyses. Alternative analyses assumed that patients without recorded information on specific adverse predictors did not have these adverse predictors; these analyses yielded similar results. Analyses using imputation for missing values also reached qualitatively similar conclusions (data not shown).
We also performed a complementary training-validation analysis (12, 13). The cohort was split randomly, and models were built in one group of patients (the training set; n = 362) and then independently validated in the other group of patients (validation set; n = 361). Patients with any identified adverse predictor were compared against patients without any such predictors by use of Kaplan-Meier plots and the log-rank test, as well as by Cox regression.
Finally, we performed a meta-analysis considering the current study and 2 previous evaluations of mortality in patients with primary SS (5, 6). The proportions of deaths attributed to lymphoma were weighted by the inverse of their fixed and random effects variances. Heterogeneity testing used Fisher's exact test.
Analyses were conducted using SPSS software, version 10.0 (SPSS, Chicago, IL). P values are 2-tailed.
- Top of page
- PATIENTS AND METHODS
Findings in the study cohort. The total followup was 4,384 person-years (mean 6.06 years, median 5.25 years, interquartile range [IQR] 2.25–9.34). The median age of the patients was 52.8 years (IQR 44–62). Other patient characteristics are shown in Table 1. The profile of the incident primary SS cases was largely similar to the profile of the whole cohort.
Table 1. Characteristics of the study patients*
|Entire study cohort, no. positive/no. with data (%)||Incident cases only, no. positive/no. with data (%)|
|Female||681/723 (94.2)||548/587 (93.4)|
|Ocular symptoms||688/720 (95.6)||558/586 (95.2)|
|Oral symptoms||677/710 (95.4)||554/580 (95.5)|
|Ocular signs (by rose bengal and/or Schirmer's test)||534/617 (86.5)||437/510 (86.5)|
|Positive histopathologic findings||673/684 (98.4)||551/561 (98.2)|
|Parotid enlargement||299/684 (44.4)||235/549 (42.8)|
|Splenomegaly||17/723 (2.4)||15/587 (2.6)|
|Lymphadenopathy||127/723 (17.6)||108/587 (18.4)|
|Palpable purpura||56/723 (7.7)||42/587 (7.2)|
|Anti-Ro antibodies||316/655 (48.2)||250/537 (46.6)|
|Anti-La antibodies||170/637 (26.7)||136/523 (26.0)|
|Antinuclear antibodies||569/707 (80.5)||464/578 (80.3)|
|Rheumatoid factor||342/662 (51.7)||279/551 (50.6)|
|Low C3 (<50 mg/dl)||17/601 (2.8)||14/507 (2.8)|
|Low C4 (<20 mg/dl)||122/601 (20.3)||92/507 (18.1)|
Thirty-four of the 723 study patients (4.7%) were lost to followup. Only 14 of these patients had been followed up for at least 6 months; the other 20 patients had made only 1 or 2 visits. Of the 14 patients with at least 6 months of followup, only 4 were older than 65 years at their last visit, and none had major diseases suggestive of imminent death. The risk of developing LPD was not different between the patients who had last been seen after July 1, 1999 at the 2 centers, those who were reached (directly or indirectly) by telephone, and those who were lost to followup (log-rank P = 0.62).
Thirty-nine deaths were recorded (8.9 per 1,000 person-years), 7 of which occurred in patients with lymphoma. The remaining deaths were due to other neoplasms (n = 10), stroke (n = 6), coronary artery disease (n = 2), sudden death (n = 1), natural death (n = 1), pulmonary embolism (n = 1), respiratory failure (n = 1), pneumonia (n = 1), sepsis (n = 1), encephalopathy (n = 1), cardiopulmonary failure (n = 1), cirrhosis (n = 1), drowning (n = 1), cardiomyopathy (n = 1), sepsis after cholecystectomy (n = 1), and unknown reasons (n = 2). Seventeen deaths occurred in patients with either LPD or cryoglobulinemia, 16 patients had neither LPD nor cryoglobulinemia, and 6 did not have any evaluation for cryoglobulins. Of the 22 patients who died before age 70 years, 11 had either LPD or cryoglobulinemia.
Of 38 patients with LPD, 14 had a diagnosis of LPD before the first study visit. The majority of the 38 cases of LPD consisted of non-Hodgkin's lymphomas (n = 30). There were no cases of Hodgkin's lymphoma. Other diagnoses were chronic lymphocytic leukemia (n = 2), isolated 25% B cell infiltration of the bone marrow (n = 1), monoclonal B cell process (n = 1), multiple myeloma (n = 1), lymphomatoid papulomatosis or T cell LPD (n = 1), and unspecified LPD (n = 2). There was a B cell origin in 34 of the 35 cases with available data. Thirteen cases were mucosa-associated lymphoid tissue lymphomas. Three patients who developed LPD had received immunosuppressive treatment (including corticosteroids [n = 2], methotrexate [n = 1], cyclophosphamide [n = 1], and azathioprine [n = 1]) prior to the LPD diagnosis.
At the time of the original diagnosis, the malignancy was high-grade in 4 patients, intermediate-grade in 1 patient, low-grade in 32 patients, and the grade had not been recorded in 1 patient. In 3 patients, low-grade lymphomas evolved into high-grade malignancies, according to the results of new biopsies performed during followup. Of the 7 deaths from lymphoma, 5 occurred in patients with documented high-grade tumors; more recent biopsy documentation was not available in the other 2 cases. The grade of the tumor at diagnosis of LPD was the strongest predictor of death (log-rank P < 0.001). The median survival time in patients with high-grade lymphoma was 3.3 years; no patient survived more than 5.2 years after such a diagnosis. The survival rates at 4 years and 8 years for non-high-grade tumors were 96.4% and 75%, respectively.
Predictors of death and LPD. The standardized mortality ratio for the study cohort was 1.15 (95% CI 0.86–1.73) compared with the age- and sex-adjusted general population of Greece. The overall risk of death was 3.6% at 5 years and 10.5% at 10 years of followup; the respective risks for developing LPD were 2.9% and 4.8%. In incident cases, 5- and 10-year mortality rates were 3.5% and 9.5%, respectively, and the rates of LPD were 2.6% and 3.9%, respectively.
In univariate age-stratified analyses, low C4 levels, splenomegaly, lymphadenopathy, palpable purpura, parotid enlargement, and the presence of rheumatoid factor were associated with increased mortality rates (Table 2). There were also trends toward increased mortality rates in prevalent cases and in the presence of antinuclear antibodies. In multivariate modeling, only low levels of C4 was retained as an independent predictor (Figure 1).
Figure 1. Kaplan-Meier plots for the risk of death in 601 patients in whom C4 levels were measured within 6 months of the beginning of followup, grouped according to low (n = 122) and normal (n = 479) levels of C4. In the remaining 122 patients, no C4 testing was done during the first 6 months of followup. The mortality rate and other important characteristics of the group of patients with missing data did not differ from those of the patients with available C4 data.
Download figure to PowerPoint
Table 2. Predictors of mortality and of the development of LPD in the entire study cohort*
| ||Predictors of mortality (n = 723)||Predictors of LPD (n = 709)|
|Hazard ratio (95% CI)||P||Hazard ratio (95% CI)||P|
|Prevalent case||2.11 (0.95–4.67)||0.068||2.52 (1.03–6.13)||0.042|
|Clinical site (loannina)||2.00 (0.65–6.09)||0.23||0.73 (0.30–1.75)||0.48|
|Male sex||2.02 (0.70–5.72)||0.20||1.58 (0.37–6.73)||0.54|
|Age (per year)||–||–||1.006 (0.974–1.039)||0.71|
|Ocular symptoms||0.87 (0.12–6.6)||0.89||0.91 (0.12–6.74)||0.93|
|Oral symptoms||0.57 (0.13–2.49)||0.45||1.23 (0.17–9.09)||0.84|
|Ocular signs (by rose bengal and/or Schirmer's test)||1.45 (0.43–4.86)||0.55||1.48 (0.34–6.34)||0.60|
|Positive histopathologic findings||NE||0.69||NE||0.68|
|Parotid enlargement||1.99 (1.01–3.91)||0.047||5.56 (1.89–16.4)||0.002|
|Splenomegaly||7.55 (2.54–22.4)||<0.001||4.09 (0.96–17.4)||0.057|
|Lymphadenopathy||2.09 (1.04–4.19)||0.038||2.62 (1.14–6.00)||0.023|
|Palpable purpura||3.00 (1.20–7.49)||0.019||5.05 (2.09–12.7)||<0.001|
|Antinuclear antibodies||2.97 (0.91–9.77)||0.072||2.56 (0.61–10.9)||0.20|
|Anti-Ro antibodies||0.97 (0.47–1.99)||0.92||3.17 (1.25–8.03)||0.015|
|Anti-La antibodies||1.20 (0.54–2.66)||0.65||2.47 (1.09–5.60)||0.030|
|Rheumatoid factor||2.53 (1.19–5.36)||0.016||1.67 (0.71–3.95)||0.24|
|Low C3 (<50 mg/dl)||1.69 (0.22–12.9)||0.62||3.90 (0.91–16.8)||0.068|
|Low C4 (<20 mg/dl)||4.39 (2.18–8.83)||<0.001||3.11 (1.32–7.31)||0.009|
In univariate analyses, the risk of developing LPD was higher in prevalent cases than in incident cases, and it increased when there was parotid enlargement, anti-Ro and anti-La antibodies, palpable purpura, and low C4 levels (Table 2). In multivariate modeling, parotid enlargement (hazard ratio [HR] 5.21, 95% CI 1.76–15.4; P = 0.003), palpable purpura (HR 4.16, 95% CI 1.65–10.5; P = 0.002), and low C4 levels (HR 2.40, 95% CI 0.99–5.83; P = 0.052) were independent predictors of LPD. Patients who had at least 1 of these 3 adverse risk factors at the onset of followup (n = 367) had a 9.08-fold higher risk of developing LPD in the future (95% CI 2.13–38.7) than those who did not (n = 342) (Figure 2). Only 2 patients developed LPD despite having none of these adverse predictors at baseline; however, C4 levels had not been measured in either of these patients.
Figure 2. Kaplan-Meier plots for the risk of developing a lymphoproliferative disease (LPD) in 709 patients without a diagnosis of LPD at the beginning of study followup, grouped according to the presence (n = 367) and absence (n = 342) of at least one adverse predictor (parotid enlargement, palpable purpura, low C4 levels) at baseline. Only 2 patients who had no adverse predictor at baseline developed LPD; C4 levels had not been measured in either patient.
Download figure to PowerPoint
Of the 7 cases of lymphoma that eventually led to death, 2 cases had been diagnosed before the beginning of the study followup. All of the 5 remaining cases had either palpable purpura or low C4 levels at the first study visit (palpable purpura in 4, and low C4 levels in 4). Low C4 levels and/or palpable purpura was present in 148 of the 709 (20.9%) patients without LPD at the first study visit.
The predictive associations were similar when the analysis was limited to incident primary SS cases. As shown in Table 3, the magnitude of the hazard ratios did not change substantially compared with the analyses that included the whole study cohort, perhaps with the exception of lymphadenopathy, which was no longer associated with death or LPD risk. In multivariate modeling, only low C4 levels were retained as a predictor of mortality, while again, palpable purpura (HR 3.96, 95% CI 1.20–13.1; P = 0.024), parotid enlargement (HR 8.35, 95% CI 1.88–37.1; P = 0.005), and low C4 levels (HR 2.72, 95% CI 0.91–8.09; P = 0.073) were the 3 selected independent predictors of LPD risk.
Table 3. Predictors of mortality and of the development of LPD among incident cases*
| ||Predictors of mortality (n = 587)||Predictors of LPD (n = 580)|
|Hazard ratio (95% CI)||P||Hazard ratio (95% CI)||P|
|Clinical site (loannina)||2.86 (0.64–12.5)||0.17||1.17 (0.37–3.74)||0.79|
|Male||2.41 (0.84–7.12)||0.10||2.09 (0.48–9.17)||0.33|
|Age (per year)||–||–||1.000 (0.963–1.038)||0.99|
|Ocular symptoms||0.59 (0.08–4.45)||0.61||0.67 (0.09–5.07)||0.70|
|Oral symptoms||0.31 (0.07–1.43)||0.13||0.86 (0.11–6.50)||0.89|
|Ocular signs (by rose bengal and/or Schirmer's test)||1.20 (0.35–4.12)||0.78||1.10 (0.25–4.86)||0.90|
|Positive histopathologic findings||NE||0.70||NE||0.72|
|Parotid enlargement||1.71 (0.81–3.59)||0.16||5.15 (1.46–18.1)||0.011|
|Splenomegaly||7.34 (2.11–25.5)||0.002||5.18 (1.18–22.7)||0.029|
|Lymphadenopathy||1.27 (0.54–2.99)||0.58||1.70 (0.60–4.83)||0.32|
|Palpable purpura||1.97 (0.57–6.77)||0.28||4.17 (1.36–12.8)||0.013|
|Antinuclear antibodies||2.70 (0.81–8.99)||0.11||1.79 (0.41–7.82)||0.44|
|Anti-Ro antibodies||0.63 (0.26–1.52)||0.30||3.54 (1.14–11.0)||0.029|
|Anti-La antibodies||0.83 (0.31–2.22)||0.71||2.79 (1.05–7.42)||0.041|
|Rheumatoid factor||2.08 (0.94–4.59)||0.071||2.07 (0.72–5.97)||0.18|
|Low C3 (<50 mg/dl)||2.30 (0.30–17.8)||0.42||2.62 (0.34–19.9)||0.35|
|Low C4 (<20 mg/dl)||2.97 (1.28–6.90)||0.011||3.03 (1.10–8.40)||0.033|
Training-validation approach. In the training set of 362 patients, low C4 levels and palpable purpura were selected as independent predictors of death in an age-stratified model. Patients with either adverse predictor were more likely to die than patients without the predictors (deaths 9 of 71 versus 15 of 291; age-stratified log-rank P = 0.0006). In the same training set, parotid enlargement and palpable purpura were selected as independent predictors of LPD. Patients with either of these adverse signs were more likely to develop LPD than patients without the adverse signs (10 of 154 versus 1 of 202; log-rank P = 0.024)
The predictive models also functioned well in the validation set of 361 patients. For mortality, there were 7 deaths among 83 patients with documented low C4 levels or palpable purpura, compared with 8 deaths among the 278 other patients (age-stratified log-rank P = 0.010). In the presence of either risk factor, the age-stratified hazard of death increased 3.76-fold (95% CI 1.22–11.5; P = 0.021) in the validation set. For LPD, 11 of 161 patients with parotid enlargement or palpable purpura developed LPD, compared with only 2 of the other 192 patients (log-rank P = 0.012). In the presence of either risk factor, the hazard of LPD increased 5.53-fold (95% CI 1.22–25.0; P = 0.026) in the validation set.
Meta-analysis for disease-attributed mortality. Two previous investigations of long-term outcomes in primary SS (5, 6) evaluated 30 and 50 patients, respectively (mean followup 10–12 years and 7.2 years, respectively). Lymphoma caused 3 of the 8 deaths in one study, but none of the 11 deaths in the other study. Merging these data with our data, we estimated that 20% of deaths were due to lymphoma (95% CI 11–34% by fixed effects and 9–40% by random effects). There was no significant between-study heterogeneity.
- Top of page
- PATIENTS AND METHODS
This study provides large-scale evidence about the long-term outcomes of patients with primary SS as well as information about how to classify the syndrome based on simple predictors identified during routine initial evaluations. Approximately 1 of 5 deaths in patients with primary SS is caused by lymphoma. This may possibly translate to a small excess in the mortality risk compared with the general population. However, the 10-year risk of developing LPD is only 4%. Moreover, all cases of subsequent lymphoma that was aggressive enough to lead to death occurred among patients who had presented with either low C4 levels or palpable purpura at their first visit for primary SS. Four of 5 patients with primary SS have neither of these adverse predictors and may be reassured of a benign course. The maximum followup of the study cohort was 18 years, but it should be noted that the mean followup was slightly over 6 years. Therefore, these conclusions should be tempered by some uncertainty when longer-term outcomes are considered.
Deaths from lymphoma occurred primarily in patients with high-grade tumors. A diagnosis of low-grade lymphoma did not carry a particularly adverse prognosis on its own, with 3 of 4 patients surviving for 8 years after such a diagnosis. The present series is the largest single-country experience on primary SS–associated LPD. Consistent with previous reports (14), we confirmed the preponderance of B cell malignancies, in particular B cell non-Hodgkin's lymphoma. More importantly, we found that simple predictors, such as low C4 levels and palpable purpura, may identify early the few patients who are likely to be at risk of dying from lymphoma.
The same predictors, along with parotid enlargement, were able to predict almost all future LPD diagnoses, regardless of their grade and aggressiveness. Early investigations had also suggested that parotid enlargement, as well as lymphadenopathy or splenomegaly, may be associated with LPD (4, 15, 16). The importance of palpable purpura and hypocomplementemia had also been proposed in an earlier analysis of a smaller number of patients from the Ioannina cohort, which had a more limited followup (7). The current experience extends these observations. We document in a large number of patients that the risk of developing LPD among patients without these easily determined adverse predictors is minimal. This was also clearly seen, when we used a training-validation approach, which further supports the generalizability of the associations (12, 13). Even more parsimonious was the validated association of death with low C4 levels.
The predictive role of low C4 levels reflects a pathogenesis component involving the activation of the classical complement pathway through immune complexes. C3 levels were not as important as C4 levels in predicting primary SS outcomes. However, this could be because decreases in C3 levels below normal are very uncommon, and thus, C3 data have insufficient statistical power to show a strong predictive effect.
Based on these data, we suggest a predictive classification of primary SS into 2 new distinct disease categories with very different clinical risks. Patients with low C4 levels and/or palpable purpura may be classified as a new high-risk disease syndrome (“type I autoimmune epitheliitis”). This group comprises ∼20% of the current primary SS diagnoses. Patients without these 2 predictors (80% of all primary SS diagnoses) may be reassured that they have a low-risk (“type II”) form of primary SS that carries no increased risk of death.
Besides LPD, patients with primary SS may also have other important morbidities, such as glomerulonephritis and peripheral neuropathy (7, 17). We did not systematically collect information about these events for the present study. In general, bias in a retrospective study may be larger for less-prominent clinical events. Nevertheless, preliminary evaluations in subsets of our cohort have suggested that glomerulonephritis and peripheral neuropathy are strongly associated with LPD (7, 17). Therefore, it is conceivable that the predictive models may apply to the development of other primary SS–associated morbidities as well (7).
Some caveats need to be acknowledged. First, this was a retrospective evaluation, and some information was missing because some patients were lost to followup. Nevertheless, losses to followup were minimal, and a few unrecorded deaths would not change the overall pattern. If 3 missed deaths among patients lost to followup are imputed, then the estimated standardized mortality ratio becomes 1.22, a value consistent with the likely magnitude of the excess mortality due to LPD. Some underestimation of the risk of LPD is also possible, allowing for some missed diagnoses due to irregular followup and for patients without overt symptoms. However, it is unlikely that these cases would represent serious cases with substantial morbidity that went completely unnoticed.
Second, diagnosis bias (18) may have been present. Patients with abnormal physical signs or laboratory results would be more likely to undergo a more extensive evaluation for the diagnosis of LPD. The outcome of death is more objective in this regard. Still, the data suggest that in the absence of low C4 levels and palpable purpura, an excess mortality risk is improbable. Thus, 4 of 5 patients with primary SS may be reassured about their benign prognosis. Extensive evaluations and specialized tests are not useful in these patients.
We should mention that cryoglobulins were also commonly seen in patients who died during this study (7, 19). However, cryoglobulins were typically measured for specific indications in our patients, and may be even more subject to diagnosis bias. Cryoglobulin levels should probably not be considered as a routine test in patients with uncomplicated primary SS, although they may be used as an adjunct in classifying a patient in the high-risk primary SS category. Cryoglobulinemia may often be the common pathway for the consumption of complement and for the clinical manifestation of palpable purpura (19), and it may reflect the pathophysiology of type I primary SS.
Finally, patients who seek medical care may differ from patients who remain undiagnosed in the community. However, patients with undiagnosed primary SS probably should not have a worse prognosis in terms of survival than those who seek medical care. Indeed, it is more likely that undiagnosed patients may have fewer serious evolving sequelae, in particular LPD, that are likely to lead to medical attention. In fact, our study included several patients with LPD who had been referred to our centers and in whom the diagnosis of the malignancy preceded the beginning of the study followup. For these patients, the prognosis should be determined primarily from the features of the LPD, rather than the features of the primary SS. Acknowledging these caveats, the present study provides large-scale evidence on the overall benign prognosis of primary SS and separates a small, but important and distinct group of high-risk patients.