Prediction of Mortality in Patients with Cystic Fibrosis

Measurement of Variables

The study cohort consisted of 673 patients with cystic fibrosis who were followed at the Hospital for Sick Children, Toronto, between 1977 and 1989. The diagnosis was based on typical clinical findings or a family history of cystic fibrosis, together with an abnormal sweat chloride value.¹ Since 1977, a broad range of clinic data have been recorded at diagnosis and at each clinic visit (every three months) and incorporated into a computerized data base.¹³ All patients over six years old undergo routine testing of pulmonary function at six-month intervals. Forced vital capacity (FVC) and forced expiratory volume in one second (FEV₁) were expressed as percentages of the normal predicted values for height and sex that have been established in our laboratory. Blood gas levels —the partial pressure of arterial oxygen (Pa0₂) and carbon dioxide (PaCO₂) — were determined in "arterialized" blood samples.¹⁴ Height and weight percentiles and weight as a percentage of the ideal weight for height and sex (weight-for-height percentage) were computed¹⁵ with use of the standards of Tanner et al.¹⁶

All patients whose pulmonary function was evaluated at least once before the end of 1987 were included in the study, thus allowing at least two years of follow-up in all patients. Records of pulmonary function and growth measurements from the date nearest each birthday in the years 1977 to 1987 were collected for each patient, along with the date of death. The percentages of patients who died within one or two years of particular measurements were calculated according to pulmonary-function variables, sex, and age. Since all the above measurements are performed routinely in patients in our clinic, it was deemed unnecessary to obtain informed consent from the patients studied or approval for the project from the Human Ethics Committee of the hospital.

Statistical Analysis

To assess the relative value of the variables in predicting death one or two years in advance, whether as independent variables or in combination with others, Cox proportional-hazards regression analysis was performed with a FORTRAN program that incorporated time-dependent covariates.¹⁷ For the purpose of this survival analysis, each patient was considered to have entered the study on the date of the first pulmonary-function test conducted in the study period. At the time of each death, values for covariates measured one or two years earlier in the patient were incorporated into a comparison, with similarly lagged values for the patients surviving at that point. The covariates included in the regression models included sex, age, FEV₁, FVC, Pa0₂, PaCO₂, and weight-for-height percentage. Regression-analysis models were produced for each covariate separately and for different combinations of pulmonary-function variables with age and sex. P values are two-tailed.

One hundred ninety patients (80 of whom were male and 110 female), or 28 percent of the 673 patients with cystic fibrosis who were followed in our clinic and underwent pulmonary-function tests, died during the 13-year study period. We excluded 16 patients who died within one year of their first measurements in the study period and 29 who did not undergo testing at our hospital within two years of their death. Twelve percent of the remaining 145 patients died before the age of 11 years, 71 percent died between 12 and 23 years of age, and 17 percent died between 24 and 44 years of age. Two thirds of the patients who died before the age of 24 years were female, whereas only one third of those who died at an older age were female.

Figure 1. Figure 1. One-Year and Two-Year Mortality Rates among Patients with Cystic Fibrosis, According to Pulmonary-Function Variable.

Values were calculated from pooled measurements. (None of the patients had Pa02 values of 76 to 80 mm Hg.)

Figure 1 shows the proportion of patients who died within 1 or 2 years, calculated from pooled annual measurements made during a 10-year period (1977 to 1987). The risk of death increased as all values for lung-function variables decreased. The approximate levels at which 50 percent of the patients died within two years of measurement were as follows: for FEV¹, below 30 percent of the predicted value; for FVC, below 40 percent of the predicted value; for Pa0₂, below 55 mm Hg; and for PaCO₂, above 50 mm Hg.

Figure 2. Figure 2. One-Year and Two-Year Mortality Rates According to Weight-for-Height Percentage (Top Panel) and Two-Year Mortality According to Weight-for-Height Percentage and Age (Bottom Panel).

Values were calculated from pooled measurements.

A decrease in the weight-for-height percentage was also associated with an increase in the mortality rate: patients with values below 70 percent had a mortality rate above 50 percent within two years (Fig. 2). As a predictor of death, this variable (an indicator of nutritional status) had no more predictive value among the male patients than among the female patients. In the younger patients (6 to 17 years old) the weight-for-height percentage had less of an association with mortality within two years than in the older patients ( 18 to 44 years old) (Fig. 2).

Table 1. Table 1. Results of Regression Analysis with Models Using a Single Covariate.

Table 2. Table 2. Results of Regression Analysis with Selected Models Using More than One Covariate.

Proportional-hazards regression analysis confirmed that all the covariates except age were significant predictors of death (Table 1). We also assessed the relative importance of the covariates in multiple-variable models (Table 2). Female sex was a significant predictor of a higher mortality rate in all models. The coefficient for age became negative and highly significant when FEV₁ or FVC was included in the model, confirming that for given values of these measurements, younger patients were at higher risk of death within one or two years than older patients. Age was not a significant factor when Pa0₂ or PaCO₂ was included in the model. The importance of age and sex was greatly reduced in the three-variable model that included weight-for-height percentage, reflecting the poorer nutritional status of female and older patients. Models with more than three predictor variables (data not shown) were also produced, but only age, sex, and FEV₁ remained significant (P<0.001) in any of these models, suggesting that this three-variable model was the most useful. In particular, including Pa0₂ had a marginal effect (P = 0.03; relative risk per change of —5, 1.1), and including PaC0₂ (P = 0.8), FVC (P = 0.2), and weight-for-height percentage (P = 0.09) had no effect.

The appropriateness of the model that included age, sex, and FEV₁ was tested further. A model that included the year of death of each patient showed that calendar time was not a significant modifier (P = 0.7). The assumption of proportionality was tested for each covariate by including a time-dependent compound variable (covariate × log [time]). There was no evidence of nonproportionality. Separate models were fitted for male and female patients, but these models did not show within-sex variations of the effects of age and FEV₁. A model was also produced from binary variables, to represent categories of FEV₁, based on increments of 10 percent of the predicted value. This model showed an approximate doubling of risk with each reduction by 10 percent of the predicted value, similar to the much simpler model in which FEV₁ served as a continuous variable.

Figure 3. Figure 3. Two-Year Mortality Rate According to FEV₁ and Sex (Top Panel) and Age (Bottom Panel).

Values were calculated from pooled measurements.

Figure 3 shows the effect of sex and age on the value of FEV₁ for predicting death within two years, as in Figure 1. The 2-year mortality rate was calculated from pooled annual measurements for the 10-year period. The risk of death among female patients with FEV₁ values between 20 and 30 percent was 56 percent, as compared with 38 percent among male patients. Likewise, children and adolescents had higher mortality rates for a given FEV₁, value than did adults (Fig. 3). Thus, the two-year mortality rate among patients less than 18 years old who had FEV₁, values that were 20 to 30 percent of the predicted value was 50 percent, as compared with 40 percent in patients 18 years old or older. The same pattern was seen when patients 6 to 17 years old were divided into those 6 to 11 years old and those 12 to 17 years old: the younger patients had higher mortality rates for a given FEV₁ value.

Our study of patients with cystic fibrosis shows that severely impaired lung function is associated with poor short-term survival. Measurement of lung function should therefore form an integral part of the routine care of patients with cystic fibrosis. It provides a more objective assessment of the extent and progress of pulmonary disease in cystic fibrosis than do clinical scoring systems, which have been used in many previous studies to predict long-term survival.^{6 , 10}

Our data show that patients with cystic fibrosis and values for FEV₁ that are less than 30 percent of the predicted value have a 50 percent chance of dying within two years. In contrast, Huang et al.¹⁰ found that lung-function measurements were not good prognostic indicators, but these investigators examined only the 5-year survival of patients after the age of 18 years. In a previous study from three U.S. clinics, Wagener et al.¹⁸ reported that the mean survival of patients with an FEV₁ less than 34 percent of the predicted value was 42 months, and only 26 months if carbon dioxide was retained (PaCO₂ >40 mm Hg).

We found that the predictive value of FEV₁ differed when we evaluated the patients according to subgroups. Female patients and patients younger than 18 years old, for example, had a less favorable prognosis for a given value of FEV₁. Although investigators at some clinics have not observed a difference in survival between male and female patients,^{9 , 10} especially among older patients,¹⁰ national data^{6 , 8 , 11 , 19 , 20} indicate that male patients have remarkably better survival than female patients. In our study, the predictive value of age and sex decreased when nutritional status, as reflected by the weight-for-height percentage, was included in the analysis. This observation supports previous suggestions that the higher mortality among females may be related in part to a concomitant decrease in nutritional status.^{12 , 13} In our study, however, nutritional status was not a good predictor of two-year mortality, particularly in the younger patients.

Relative-risk estimates provide a quantitative method to describe the increased risk of death associated with each clinical characteristic. For example, at a given age and level of FEV₁, female patients have a risk of death within two years more than double that of male patients. In patients of both sexes of any age, a decline in the FEV₁ of 10 percent of the predicted value over a short time is associated with a twofold increase in risk, and of two patients of the same sex who have the same FEV₁, level, the patient who is 10 years younger has a two-fold greater risk.

Although the FEV₁ was found to be the most significant predictor of survival in our study, it should be pointed out that lung function m individual patients with cystic fibrosis can be influenced by several factors (e.g., sputum colonization with pseudomonal species, pancreatic function, and the presence of increased airway reactivity, atopy, and passive and active exposure to smoke).^{6 , 10} Furthermore, previous studies have shown that other factors not included in our Cox models, such as the mode of presentation,⁹ the extent of lung disease at diagnosis,⁹ treatment regimens,¹¹ and psychosocial status,⁸ can influence survival.

In view of the patterns of survival presented, we believe that patients with FEV₁ values less than 30 percent of the predicted value should be considered candidates for lung transplantation. Female patients and patients under 18 years old may need to be considered for lung transplantation at an earlier stage. It should be stressed, however, that the final decision to proceed with transplantation should be based on a comprehensive assessment of the patient's overall medical and psychosocial condition.⁵

There are considerable geographic differences in the survival of patients with cystic fibrosis^{6 7 8 9 10 11 12 13}; patients in our clinic generally have a more favorable prognosis than their counterparts in other North American clinics and elsewhere.^{6 , 12 , 13} Can the results of our study therefore be extrapolated to experiences at other centers? During the past 20 years the survival rates of patients cared for at our clinic have plateaued, while the corresponding rates at other clinics in North America and in foreign countries have been steadily improving.^{6 7 8 9 10 11 , 20} The net result has been a more uniform survival among patients from different geographic locations. Thus, the median survival of patients attending 115 clinics for cystic fibrosis in the United States in 1990 was 27.6 years (Fitzsimmons S: personal communication), as compared with a median survival of 30.9 years among patients attending 32 specialized Canadian clinics between 1985 and 1989.¹⁹ Therefore, we think that our findings should enable care givers at other clinics to identify suitable candidates for lung transplantation. In this regard, patients with cystic fibrosis who were placed on an active waiting list at two British transplantation centers had values for FEV₁ similar to those of patients who, on the basis of our study, would also be recommended as candidates.⁵

The principal barrier to lung transplantation is the limited availability of donors, although legislative initiatives^{21 , 22} and better techniques of lung preservation²³ may improve the situation. In addition, lung transplantation remains an expensive procedure, although less expensive than the medical treatment of cystic fibrosis during its terminal phase.² It is hoped that our data will make it possible to select the most appropriate patients for lung transplantation, and that meticulous attention to their medical and nutritional needs will optimize their likelihood of benefiting from this limited resource.