INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The interstitial lung diseases are a group of chronic pulmonary disorders that result from derangements of the alveolar walls and loss of alveolar-capillary units (1). Diffuse infiltrative lung disease is probably a better descriptive term because there is often involvement of alveolar air spaces and distal airways as well. However, in referring to such a condition, most pulmonologists prefer the term interstitial lung disease (ILD) despite its limitations.

A previous retrospective study of 48 children with ILD demonstrated that the spectrum of pediatric ILD includes a large, heterogeneous group of rare disorders (4). Despite diversity, these diseases have in common features of restrictive lung disease and disordered gas exchange that are associated with high morbidity and mortality.

To date, almost no information exists about factors that might be associated with outcome in these children. Therefore, we performed the following study to investigate factors associated with survival for children with ILD, including certain clinical measures of chronic lung disease.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study was part of a larger descriptive, observational investigation of the diagnosis and outcome of pediatric ILD. The study population included children with ILD evaluated at the University of Colorado-affiliated hospitals (University Hospital, Children's Hospital, and National Jewish Center for Immunology and Respiratory Medicine) over a 15-yr period (1980 to 1994). Specific clinical information relating to the history, physical examination, diagnostic evaluation, treatment, and outcome of these patients was obtained and systematically recorded in a data base. The data were collected retrospectively from a chart review of patients seen between 1980 and 1991 (Group 1), and prospectively for new patients evaluated between 1992 and 1994 (Group 2). A clinical description of the patients in Group 1 has been previously reported (4). The data for both groups were compared so as to be sure that there were no significant baseline differences, and the groups were then combined to obtain a larger sample size.

The protocol for both groups was designed to permit the study of patients without an established diagnosis, from the perspective of a clinician who, faced with a large differential diagnosis, would want to know the value of different diagnostic studies in making a diagnosis and the relevance of different outcome measures in determining prognosis. History and physical examination, noninvasive studies, and invasive studies all contributed to the diagnosis of these patients' disease, as has been reported for Group 2 patients (5). Patients 1 mo through 18 yr of age were included if they had: (1) any symptom of cough, tachypnea, rales, exercise intolerance, or hypoxemia; (2) duration of illness of at least 1 mo; and (3) diffuse infiltrates on chest film. Patients were excluded if, at the time of initial evaluation, they were known to have: (1) bronchopulmonary dysplasia; (2) congenital heart disease; (3) primary malignancy; (4) primary immunodeficiency; (5) primary autoimmune disorder; (6) an underlying condition that would predispose to aspiration; (7) cystic fibrosis; or (8) acquired immunedeficiency syndrome (AIDS). Patients were included if one of these conditions was subsequently discovered. Patients with primary pulmonary vascular diseases such as pulmonary hemangiomatosis, pulmonary vein atresia/stenosis, and venoocclusive disease were included if they presented with diffuse infiltrates of unknown origin, and one of these diseases was subsequently discovered with evaluation.

Final diagnoses were classified as specific, suggestive, or no specific diagnosis. A specific diagnosis was assigned to those cases in which a diagnosis was made with confidence based on careful consideration of clinical presentation, laboratory evaluations, and invasive diagnostic studies, such as bronchoalveolar lavage (BAL) and lung biopsy. A suggestive diagnosis was assigned to those cases in which the clinical presentation suggested a disorder (such as hypersensitivity pneumonitis from bird exposure) that could not be confirmed conclusively by diagnostic evaluation. No specific diagnosis was assigned to those cases in which no specific or suggestive diagnosis could be made despite complete diagnostic evaluation.

At the time of enrollment, one of the investigators (L.L.F.) assigned to each Group 2 patient a severity-of-illness score derived from findings in the initial evaluation. This score, based on symptoms, increasing levels of hypoxemia, and pulmonary hypertension, has been suggested previously as a means of categorizing severity of illness in children with ILD (Table 1) (1). Hypoxemia was defined in terms of an oxygen saturation (Sa_O2) of less than 90% in room air in Denver, Colorado (elevation: 1700 m), measured with a pulse oximeter. Pulmonary hypertension was defined according to standard criteria for right ventricular hypertrophy or dilatation, measured electrocardiographically or echocardiographically, and/or by elevation of pulmonary artery pressure (Ppa; mean pressure: > 30 mm) as measured through cardiac catheterization. In Group 1 patients, the severity-of-illness score was assigned by the same investigator, based on information in the patient records at the time of initial evaluation.

For this study, we extracted specific information from the data bases of Group 1 and Group 2 patients, including age of onset; duration of symptoms prior to evaluation; family history of ILD; the presence or absence of a weight below the fifth percentile for the patient's age, crackles, and clubbing on physical examination; the patient's age at the time of most recent evaluation or death; severity-of-illness score at the time of initial evaluation; and diagnosis.

Statistical Analysis

Because two methods were used for patient recruitment, the patient groups were first compared to assess possible baseline differences. Categorical variables used in these analyses were tested with chi-square analysis, and the t test was used for continuous variables.

Kaplan-Meier estimates of mean and median survival times (6) were derived from the product-limit survival estimate generated by the use of SAS for Windows, release 6.11 (Statistical Analysis Systems; SAS Institute, Cary, NC). The origin of time in survival analyses was defined as the age at onset of symptoms of ILD. Age at onset of symptoms marked the start of the period for which each patient was at risk of death specifically due to ILD. To control for possible selection bias (i.e., patients' deaths are not observable before they are enrolled in a study), left truncation was used in the survival analyses. A time-dependent covariate was entered into the models to control for times at which a patient was not in the risk set. Survival was calculated to the date of death (uncensored) or to the date of the last follow-up visit (censored). The log-rank test (7) was used to test for statistically significant group differences in survival.

Cox proportional hazards modeling was employed to evaluate the effects of several variables on survival, including clinical features and age of onset of ILD symptoms (8). The univariate statistical significance of each variable was determined with the maximum likelihood estimator of the regression coefficients. The most highly significant variable was selected first, and the remaining variables were entered into the model to identify additional significant variables. At each step, the proportional hazards assumption was tested. This procedure was repeated in a stepwise fashion until no additional statistically significant variables were identified.

To compare the survival of patients with varying symptom durations, symptom duration was dichotomized for the Kaplan-Meier analysis (i.e.,  1 yr versus > 1 yr). Despite this, symptom duration was entered as a continuous variable in the proportional hazards modeling.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Ninety-nine patients (48 in Group 1 and 51 in Group 2) met the entry criteria for the study. The larger number of patients in the prospective study reflects both the difficulty in identifying cases retrospectively and our growing interest in and awareness of pediatric ILD in recent years. This is supported by the finding that 9, 14, 25, and 51 patients were initially evaluated in the years, 1980 through 1983, 1984 through 1987, 1988 through 1991, and 1992 through 1994, respectively.

Patient Characteristics

Initial analyses indicated that the frequency distributions for selected clinical features documented at the time of initial evaluation for ILD did not differ significantly between the two groups of patients. These factors included a weight of less than the fifth percentile for age, crackles, clubbing, duration of symptoms exceeding 1 yr, a family history of ILD, and severity-of-illness score (Table 2). The mean ages ± SD (months) at onset of ILD also did not differ significantly between the two groups (43.5 ± 67 mo; range: 0 to 212 mo, versus 49.9 ± 58.5 mo; range: 0 to 186 mo; p = 0.61). On the basis of these results, the patient groups were combined for the remaining analyses.

Overall Outcome

There were 15 deaths among the 99 patients. The distribution of the deaths among the various ILD diagnoses is presented in Table 3. For the total group of ILD patients, mean survival from age of onset, corrected for initial missing observational periods of risk, was 47 mo (SE: 2.4 mo). Probabilities that a patient survived 24 mo, 48 mo, and 60 mo after onset of symptoms (corrected for missed observational periods of risk) were 83%, 72%, and 64%, respectively. Age at onset of symptoms, a quantitative covariate, had no effect on the overall survival time (p = 0.98). For each 1-mo increase in age at onset of illness, the hazard of death increased by an estimate 0.01%.

Clinical Features and Outcome

The survival probabilities, obtained with Kaplan-Meier survival analysis, for patients categorized according to selected clinical features (weight below fifth percentile for age, crackles, clubbing, family history of ILD, symptom duration at initial evaluation) did not differ statistically between the groups (Table 4).

Patients with lower severity-of-illness scores had significantly greater survival probabilities. As shown in Table 5, survival differences between patients assigned lower score(s) and patients assigned higher score(s) were statistically significant. There was also a marked correlation of decreasing probability of survival with increasing severity-of-illness score. These analyses demonstrate the utility of the severity-of-illness score for categorizing patients into discernible outcome groups based on survival.

Diagnostic Subgroups

Since nine of the 15 deaths occurred in patients with diagnosed pulmonary vascular disease (PVD) or desquamative interstitial pneumonitis (DIP), analyses were conducted to control for the effect of those two diagnostic subgroups on outcome. As expected, the probability of survival without including the PVD or DIP patients was 83% at 48 mo. This patient group, however, had little effect on the statistical assessments of correlation of clinical features and severity-of-illness score with outcome. Symptom duration was not associated with the probability of survival, but a low severity-of-illness score remained a strong indicator of survival.

Cox Proportional Hazards Modeling

Cox proportional hazards modeling was performed to investigate the effect of several potentially important variables on overall survival of ILD patients. The final model revealed the significance of the severity-of-illness score and the diagnosis of PVD or DIP in influencing the survival of ILD patients. For each one-unit increase in the score, the risk of death increased by an estimated 140%. The risk of death for patients with DIP or PVD was increased by 680%. No other clinical features were independent prognostic factors for survival.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The findings in our study are consistent with previous reports that pediatric ILD is associated with high mortality. In a series of 25 children with fibrosing alveolitis or DIP, Sharief and colleagues (9) reported a poor response to treatment in nine patients, with four deaths. In a review of 28 patients with DIP, Stillwell and associates (10) reported that only 17 patients survived. In a series of 17 children with a variety of ILDs, including DIP, lymphocytic interstitial pneumonitis, pulmonary hemosiderosis, hypersensitivity pneumonitis, and interstitial fibrosis associated with severe combined immunodeficiency (SCD), Diaz and Bowman (11) reported that 50% had a poor quality of life and that 19% died. Osika and coworkers (12) reported nine deaths in 10 infants with idiopathic pulmonary fibrosis (IPF).

We have demonstrated that a severity-of-illness score calculated at the time of initial evaluation for patients with a wide variety of ILDs correlates with survival and may therefore be useful as a prognostic factor. This correlation proved valid even when controlling for the two diagnostic categories (PVD and DIP), that accounted for more than half the deaths in our study. Although it is widely accepted that increasing levels of hypoxemia would indicate more severe disease, and that adults with IPF who develop pulmonary hypertension have a poor prognosis (13), the relationship between these findings and outcome has never been previously investigated in children with ILD. As a clinical tool, a score based on oxygen saturation has appeal because pulse oximetry is simple, noninvasive, and reasonably accurate in both infants and older children (14). In contrast, pulmonary function testing as a measure of outcome in this population would be difficult because no measurement of pulmonary function can be easily applied to both infants and children.

To our knowledge, the population examined in the present study is the largest population of pediatric ILD patients to be reported to date, although we had to combine patients from the retrospective and prospective portions of the study to obtain a population sample large enough to analyze. However, even after we combined the groups, the small number of deaths may still have limited our power in detecting all prognostic factors influencing survival. Because of the retrospective component of the study, we were unable to determine whether changes in severity-of-illness score over time would correlate with disease progression or response to therapy.

Although familial DIP has been highly lethal in previously reported cases (including our own patients) (4, 15), our data suggest that a family history of ILD in children with other types of ILD is not associated with decreased survival. Nor in the present study were age of onset, a weight below the fifth percentile, clubbing, crackles, or duration of illness associated with decreased survival.

In conclusion, this study demonstrates that children with ILD have decreased probabilities of survival that can be predicted with a severity-of-illness score. This score may be useful as a measure of outcome, but further investigation is needed to determine whether changes in the score over time correlate with disease progression or improvement with treatment. Owing to the rarity of pediatric ILD, multicenter and international collaboration, as suggested by Hilman (18), should be encouraged to validate or extend these observations.