- Email: [email protected]
The CF-ABLE Score
CHEST
Original Research GENETIC AND DEVELOPMENTAL DISORDERS
The CF-ABLE Score A Novel Clinical Prediction Rule for Prognosis in Patients With Cystic Fibrosis Cormac McCarthy, MB; Borislav D. Dimitrov, PhD; Imran J. Meurling, MB; Cedric Gunaratnam, MB; and Noel G. McElvaney, MB
Background: Determining prognosis and predicting outcomes in cystic fibrosis (CF) is a complex issue, and there have been very few clinically applicable models for this. The aim was to create a simple, practical outcome prediction tool for CF. Methods: Forty-nine consecutive patients with CF from a single center were studied over an 84-month period (2004-2010). All baseline clinical parameters were gathered, and FEV1 measurements were analyzed over the study period. Using patterns of FEV1 decline, a tipping point of 52.8% predicted was identified. Other clinical variables were analyzed and correlated with outcome. Poor outcome was defined as death or transplantation. Using age, BMI, lung function (ie, FEV1), and number of exacerbations in the past 3 months, the CF-ABLE score was created. The score was validated for data from 370 patients from the national Cystic Fibrosis Registry of Ireland. Results: The ABLE score uses clinical parameters that are measured at every clinic visit and scored on a scale from 0 to 7. If FEV1 is , 52%, then 3.5 points are added; if the number of exacerbations in the past 3 months is . 1, then 1.5 points are added; if BMI is , 20.1 kg/m2 or age , 24 years, each receive 1 point. Conclusions: Patients with a low score have a very low risk of death or lung transplantation within 4 years; however, as the score increases, the risk significantly increases. Patients who score . 5 points have a 26% risk of poor outcome within 4 years. This score is simple and applicable and better CHEST 2013; 143(5):1358–1364 predicts outcome than FEV1 alone. Abbreviations: CF 5 cystic fibrosis; CFRI 5 Cystic Fibrosis Registry of Ireland; ROC 5 receiver operating characteristic
fibrosis (CF) is a multisystem disorder with a Cystic significantly shortened life expectancy. The major
cause of mortality in CF is lung disease.1 As the condition worsens over time, lung function declines. This decline usually is measured by FEV1.2 Previous studies suggested that patients with CF with an earlier mortality demonstrate a greater rate of decline in FEV1% predicted compared to patients with CF with Manuscript received August 14, 2012; revision accepted October 17, 2012. Affiliations: From the Respiratory Research Division (Drs McCarthy, Meurling, Gunaratnam, and McElvaney), Department of Medicine, and Department of General Practice (Dr Dimitrov), Royal College of Surgeons in Ireland, Dublin, Ireland; Academic Unit of Primary Care and Population Sciences (Dr Dimitrov), University of Southampton, Southampton, England; and Department of Respiratory Medicine (Drs McCarthy, Meurling, Gunaratnam, and McElvaney), Beaumont Hospital, Dublin, Ireland. Funding/Support: The authors have reported to CHEST that no funding was received for this study.
later mortality/longer survival.3 In this regard, decrements in lung function are associated with recurrent pulmonary exacerbations, which are common in CF, and frequent exacerbations lead to an increased rate of decline in FEV1.4,5 Decreased FEV1 at a young age has also been shown to correlate with a poor outcome.6 Many studies investigated the accuracy and usefulness of prediction tools in CF. Both clinical findings and lung function parameters have been used to create prediction models for pulmonary outcomes.7 Most have been hampered by complexity and difficulty in Correspondence to: Cormac McCarthy, MB, Respiratory Research Division, Department of Medicine, Royal College of Surgeons in Ireland, Education and Research Centre, Beaumont Hospital, Beaumont Rd, Dublin 9, Ireland; e-mail: [email protected] © 2013 American College of Chest Physicians. Reproduction of this article is prohibited without written permission from the American College of Chest Physicians. See online for more details. DOI: 10.1378/chest.12-2022
1358
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
application.8,9 This complexity has precluded these prediction tools from being used routinely, and hence, many clinical decisions are guided by FEV1 alone. A better and more practical prediction tool is needed to stratify risk and predict outcome earlier to facilitate interventions or to guide transplant assessment at an earlier stage.10 The aim of the present study was to create an easy-to-use scoring tool that predicts risk over a significant time period using commonly measured clinical parameters. The ideal prediction tool should be one that is simple and quick to score and uses relevant clinical information that can be addressed to improve outcome if necessary. Because BMI has been shown to correlate with increased mortality risk, it was also assessed in the prediction algorithm.11,12 The prediction tool was validated in a larger cohort from the national Cystic Fibrosis Registry of Ireland (CFRI). Materials and Methods Study Design and Selection of Patients This was a prospective, single-cohort study for the derivation and validation of a new clinical prediction rule, the CF-ABLE score. Approval for this study was obtained from the ethics committee of Beaumont Hospital (approval number 12/80). Initial data were obtained from a single specialist adult CF center. A cohort group of 49 consecutive patients were included in the derivation part of the study. Data were collected on pancreatic insufficiency, diabetes mellitus status, Pseudomonas aeruginosa colonization, Staphylococcus aureus colonization, genotype, FEV1, BMI, bone mineral density, and number of exacerbations. Patient FEV1 measurements from 2004 to 2009 were obtained from medical records, and all data on pulmonary exacerbations for this period were evaluated at the end of 2009. Exacerbations were defined as events in which the patient’s clinical condition changed significantly enough to require treatment with IV antibiotics administered either in the hospital or in the home setting. BMI was recorded, and initial BMI was the first measurement noted in the period. Patient age was recorded at the time of the first measurements made in the study period. At the end of 2010, the patients were retrospectively classified according to outcome, which was defined as good or poor outcome. Patients were classified with poor outcome if they were referred for lung transplantation or had died in the interim. Criteria for lung transplantation referral were a baseline FEV1 , 30% predicted, oxygen saturation , 90% at rest or on exertion, or both. Following identification of patterns of lung function decline and predictors of poor outcome in the initial cohort and derivation of the CF-ABLE rule, new data from the national CFRI were collected for validation purposes. We gathered the data for all available patients from the entire registry from 2005 to 2010 with the exception of the 49 already analyzed in the derivation cohort. Patients were excluded if they were aged , 16 years, if they had already received a double lung transplantation prior to 2005, or if no pulmonary function or BMI data were available prior to their transplant. Statistical Analysis Because this was a pilot study, a sample size was not calculated a priori; however, a post hoc calculation of power was performed,
as appropriate. Data are presented as mean ⫾ SD for continuous variables or number and frequency (percentage, proportion) for categorical variables and analyzed with two-tailed parametric (t) or nonparametric (x2, Mann-Whitney) tests, as appropriate. The analyses at group (population) level used parametric regression modeling with fitting of a number of linear and nonlinear equations to analyze and project the group FEV1 percent predicted by a range of built-in linear, cubic, polynomial, and other high-order functions (Fig 1). Logistic regression analysis was applied by entry and backward stepwise methods, with adjustment for covariate effects for variables significantly associated with poor prognosis in the univariable analyses or found clinically relevant and, eventually, important. Survival curves were based on Kaplan-Meier estimates. To test the independence over time of the predicting variables from the best logistic regression models, the cumulative hazard function of the primary end point was calculated by means of Cox regression modeling, with adjustment for covariate effects. To confirm the predictive performance of the best model, we further applied a receiving operator characteristic (ROC) curve analysis on the probabilities derived from the best model, identifying the best cutoff point for each univariate, statistically significant predictor. The statistical significance of all tests was assumed at P , .05, unless stated otherwise. All evaluations were done with SPSS version 18 (International Business Machines Corporation) statistical software. Initially, to explore various hypotheses about possible predictors of CF prognosis, we collected all pulmonary function data for patients over a 6-year period (2004-2009), which were transformed into single values for equidistant 3-month periods as means of all valid measurements for each patient. Pulmonary exacerbations were calculated as a total number for the period and subcategorized as number of exacerbations per 3-month period. All demographic and clinical variables are described and presented by their main statistical parameters, overall, and by the type of outcome. Algorithm for Derivation of CF-ABLE Score Univariate and multifactorial logistic regression analyses were applied to identify and select the significant independent predictors. Univariate analyses and ROC curves were used to identify the best cutoff points in terms of individual outcome for each predictor (e-Fig 1). Predictors were entered into a multifactor regression model with other potential predictors. The predictive performance of the best model and its individual predictors were tested by ROC curve analysis, where best cutoff points were established for each predictor according to the results and clinical judgment. The best cutoff levels were used to create new dichotomized or trichotomized variables from each single or combined independent predictor to specify its initial scores or high-risk and low-risk levels (ie, presence [1] or absence [0] of risk). Values were dichotomized or trichotomized to create a simple clinical prediction score. Subsequently, a new logistic regression model with these dichotomized/trichotomized values was constructed (e-Appendix 1, section I). The distribution of outcome according to the different levels of the CF-ABLE score as well as its predictive performance by ROC curve analysis was also computed (e-Fig 2). After the construction of the CF-ABLE score in the derivation cohort, the same parameters were taken from the national data set, and the score was further validated in the CFRI.
Results The total number of patients from the initial center was 49; 36 were classified as having had a good outcome at the end of the study in 2010 (ie, none of
journal.publications.chestnet.org
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
CHEST / 143 / 5 / MAY 2013
1359
Figure 1. Model of lung function decline in the derivation cohort. A, Demonstrates mean FEV1 percent predicted for the entire cohort plotted against time. All FEV1 measurements were transformed into single values of means for equidistant 3-month periods. These are plotted against time to demonstrate the overall pattern of decline in lung function in the derivation cohort as a group. B, Parametric regression modeling was used to analyze the pattern of FEV1 decline at the group level. Fitting of a number of linear and nonlinear equations to analyze and project the decline was done by a range of built-in linear, cubic, polynomial, and other high-order functions. There was a significant nonlinear tendency of decline in mean FEV1. The most statistically significant models of decline were a cubic equation (red curve) (P , .05), showing an apparent plateau phase and a cyclic pattern. Additionally, an inverse equation (blue curve) (P , .05) showed a rapid decline followed by a prolonged plateau phase.
these patients died or were referred for lung transplantation). They had an average initial FEV1 of 82.91% and a mean number of 1.27 exacerbations per 3-month period. Thirteen patients were classified as having had a poor outcome, of whom 10 were referred for lung transplantation (two died before 2010) and two
others received lung transplantation in 2010. Three other patients died before 2010, none of whom were referred for lung transplantation. The poor-outcome group had a lower average initial FEV1 of 39.58% and a higher mean number of exacerbations per 3-month period of 1.54 (Table 1).
1360
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
Table 1—Demographics of the Derivation and Validation Cohorts Clinical and Demographic Parameter Derivation Cohort No. patients Patients who died Patients who received or were referred for transplant Patients with poor outcome (ie, died/ transplantation) Age at first FEV1, y Male sex Female sex DF508 homozygous Initial BMI, kg/m2 Initial FEV1, % predicted No. exacerbations in initial 3 months Range
Validation Cohort
49 5 (10.2) 10 (20.4)
370 23 (6.2) 11 (2.9)
13 (26.5)
34 (9)
21.24 (20.13-22.35) 24.27 (23.55-24.99) 31 (63.27) 211 (57) 18 (36.73) 159 (43) 27 (55) 212 (57.3) 20.41 (19.39-21.47) 21.38 (20.98-21.78) 62.05 (54.76-69.34) 60.16 (57.62-62.7) 0.36 (0.12-0.54) 0.34 (0.281-0.398) 0-4
0-3
Data are presented as No. (%) or mean (95% CI). DF508 5 mutation of the CFTR gene.
The analysis of all FEV1 measurements over the 6-year period showed a significant, nonlinear tendency of decline in mean FEV1 over time (P , .05) (Fig 1A) that was indicative of a tipping point. The mean estimates from a generalized linear model were best described by both a cubic equation (P , .05) (Fig 1B) showing an apparent plateau phase in lung function (with a possible cyclic pattern) and an inverse equation (P , .05) (Fig 1B) showing a rapid decline followed by a prolonged plateau phase in FEV1 decline. The modeling also identified a parallel, nonlinear increase in the mean number of exacerbations (per 3-month period) over time (P , .05) that correlated significantly with the decline in lung function (P , .05) for all patients. The poor-outcome group had a significantly greater number of exacerbations throughout the initial 4-year period (2004-2007) compared with the good-outcome group (P 5 .02) and had a significantly greater mean number of exacerbations in the initial 3-month period. The poor-outcome group had a mean number of exacerbations of two per 3-month period initially and an average of 1.54 per 3-month period over the 4 years; the mean exacerbation rate in the good-outcome group was 1.27 per 3-month (Fig 2). The mean initial BMI of the poor-outcome group was 19.6 kg/m2, and the mean age was 21.89 years; BMI and age of the goodoutcome group were 20.63 kg/m2 and 21.01 years. Given the results from the analyses at the group level (Fig 1B, e-Figs 3, 4), we used the regression analysis and ROC curve approach to identify a value around the average FEV1 of 56.75% predicted for the good-outcome group, which served as the cutoff point to distinguish patients in this group from those
Figure 2. Mean number of exacerbations per 3-month period in the derivation cohort. The mean number of exacerbations per 3-month interval in good outcome and poor outcome throughout the initial 4-year period (2004-2007) are shown. The pooroutcome group had a greater number of exacerbations and more clustering.
in the poor-outcome group. Through a ROC curve analysis (e-Fig 1), we identified an FEV1 cutoff of 52% to 53% for initial FEV1 measurements and a similar value for mean FEV1 in the second quarter (52.8%). Patients with an FEV1 above this level had a significantly lower risk of worse outcome in the following 4 years. Derivation of the CF-ABLE Score The main univariate analysis and ROC curve identified a best cutoff point in terms of individual outcome (e-Fig 1). This main predictor (mean FEV1% predicted during the first 3 months) was then entered into a multifactor regression model with other potential predictors (BMI, age at first FEV1% predicted, mean number of exacerbations during first 3 months) (e-Appendix 1, sections I, II, III, e-Table 1) The predictive performance of the best model and its individual predictors were tested by ROC curve analysis, where best cutoff points were established for each predictor according to the results and clinical judgment in order to incorporate the four variables as predictive components of a new, simplified score. The new rule to predict poor outcome was named the ABLE score according to its components (age, BMI, lung function [ie, FEV1% predicted], and exacerbations) and constructed by the algorithm described next. The best cutoff levels were used to create new dichotomized or trichotomized variables from each single or combined independent predictor in order to specify its initial scores or high-risk and low-risk levels (ie, presence [1] or absence [0] of risk ) (e-Appendix 1, section I). The next step was to create a new logistic regression model with the dichotomized and trichotomized values (e-Appendix 1, section I). Using this methodology, the ABLE score is constructed from the sum
journal.publications.chestnet.org
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
CHEST / 143 / 5 / MAY 2013
1361
of the individual component scores and has the following ranges: ABLE score [0.0 2 7.0] 5 FEV1% predicted (mean previous 3 months) [0.0 2 3.5] 1 number exacerbations (previous 3 months) [0.0 2 1.5] 1 age at time of assessment 3 BMI [0.0 2 2.0] Validation of the CF-ABLE Score in the National CFRI Cohort During the period of 2005 to 2010, data from 370 patients from the national CFRI were included in the analysis. Twenty-three patients died and 11 patients underwent double lung transplantation for an event rate of 9.19% for the poor-outcome group (Table 1). The application of the new clinical prediction rule in the national CFRI validation cohort ranged from 0 to 7 (2.84 ⫾ 2.22). There was a significant difference between the mean ABLE scores of the two subgroups by outcome (good outcome, 2.61 ⫾ 2.15; poor outcome, 5.21 ⫾ 1.30; P , .001). The histograms of the distributions were also computed and clearly illustrated the statistically significant difference (Fig 3). The discriminative performance was further confirmed by a new ROC curve analysis (e-Fig 5), indicating a significant area under the ROC curve of 0.82 (95% CI, 0.77-0.88). The logistic regression model to predict
death or referral for transplantation in the validation cohort indicated that, on average, a 1-point increase of ABLE score relative to the previous score would lead to an 81% increase in risk (OR, 1.81; 95% CI, 1.46-2.22); for example, an increase in ABLE score from 3.50 to 4.50 would lead to an increased risk of 6.45% to 14.71%. In terms of calibration performance, it can be seen that with the increase of ABLE score, the risk increases linearly along the whole range of the score (Fig 4A, e-Appendix 1, section IV; e-Table 2, 3; e-Fig 6). Discussion We developed an easy-to-use predictive score for determining outcome in CF. This score accurately predicts the risk of death or transplantation within 4 years, as validated in the national CFRI. Previous studies identified an FEV1 , 30% predicted to be a predictor of mortality and have suggested transplant referral at this point.6 Others argued that this value is not predictive of outcome and that rate of lung function decline is more sensitive.13 In the present study, we demonstrate that there is a nonlinear decline in FEV1% predicted over time, which follows one of two patterns: either a sharp decline followed by plateau that declines further in a cyclical manner or a sharp decline followed by a prolonged
Figure 3. Histograms of distribution of the cystic fibrosis (CF)-ABLE score within both outcome groups in the validation cohort. Application of the ABLE score in the validation cohort demonstrated a statistically significant difference in the mean ABLE scores between the good- and poor-outcome groups (P , .001). The mean ⫾ SD CF-ABLE score in the good-outcome group was 2.61 ⫾ 2.15 and in the poor-outcome group, 5.21 ⫾ 1.3. 1362
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
Figure 4. A, Risk of poor outcome according to CF-ABLE score in the validation cohort. There is a linear correlation in risk for a poor outcome (death or transplantation) within 4 years as ABLE score increases. In the validation cohort, there was a significant change in risk as the ABLE score increased to . 3.5. There is a 26% chance of poor outcome within 4 years if the ABLE score is ⱖ 5. B, CF-ABLE scoring table. The ABLE score is based on adding the values for each variable of the score. The total is from 0 to 7. As the score increases, there is a linear increase in risk of poor outcome. See Figure 3 legend for expansion of abbreviation.
plateau phase.14 We identify that an FEV1 of 52% predicted is an accurate predictor of outcome and identifies high-risk patients earlier. We demonstrate that in the poor-outcome group, there was a significantly higher number of exacerbations per 3-month period over the course of the study and that the clustering of exacerbations has a deleterious effect on long-term lung function measurements and survival.4,5,15-17 Thus, the number of exacerbations is another key component of our predictive model. Finally, we identify age as having a significant effect on risk prediction, especially in patients with a low FEV1 or low BMI at a young age. This further underscores the need for goal-driven nutritional strategies to improve survival in patients with CF. Several prediction models and scoring tools have been designed for use in CF; however, most have been either too complex to score, cumbersome to use, or unreliable. For example, the Shwachman-Kulczycki score relies heavily on subjective assessments, requires a combination of radiographic and clinical parameters, is difficult to score, and is not acceptably reliable.18,19 Similarly, the Cooperman and National Institutes of Health scores are subjective and overly complex and have not been validated.9,20 There are few validated prediction rules in CF. Liou et al8 created a survivorship
model from US CF registry data. This is an efficient prediction model and has been validated in a large cohort of registry patients, but it is somewhat limited by deriving the prediction rule from the same cohort as the validation group. Additionally, the scoring system with eight parameters and at least 15 steps in the computation make it complex and difficult to use. In comparison, the CF-ABLE score has only four parameters and is applied by using very simple rules and a straightforward arithmetic calculation (Fig 4B). Furthermore, the score by Liou et al8 has been validated only by Hosmer-Lemeshow test, and no ROC analyses or any other accuracy measures were shown, unlike with the CF-ABLE score. There are several potential limitations to the present study. We had a small derivation population, but when compared with the national CFRI the patients had similar outcomes and demographics. The CF-ABLE scores in the derivation and validation cohorts were very comparable, and regarding the slight differences noted between the groups, there are a number of reasons for this. First, a higher number of events were captured in the small cohort of patients from the derivation group because this group was closely evaluated over the period of study. Second, referral for transplant was used as a proxy measure for transplantation in the derivation cohort as a clinical outcome. Third, the data collected for the validation cohort were limited by the number of patients in the registry who were not included in the study because they had undergone transplantation but had no data available in the registry prior to their transplantation. Finally, a potential problem with this scoring system is the difficulty in defining a pulmonary exacerbation in CF. We used IV antibiotics treatment at home or in the hospital as definition for a significant acute exacerbation, which may have underestimated the total number of exacerbations; however, we are confident that we identified all the severe exacerbations. This study adds to the field of stratifying patient risks at earlier time points to identify the need for earlier intervention with new treatments or alternative ways of delivering the best available treatment. It may direct CF clinicians to concentrate their therapeutic efforts on aspects of care most amenable to change. The CF-ABLE score identifies both highand low-risk patients over a prolonged period. It demonstrates that those with a low score (eg, , 2) have a small risk of death or transplantation in the next 4 years, and this should encourage clinicians and their patients to maintain care and CF-ABLE score parameters. If the patient has a high score (eg, 5 of 7), he or she has a 26% risk of death or need for a lung transplantation in the next 4 years. We believe that this tool is a robust and effective means to assess risk in CF. It has been validated in a
journal.publications.chestnet.org
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
CHEST / 143 / 5 / MAY 2013
1363
large CF population with the common CF mutations. As such, we believe that it is generalized beyond Ireland, and we would welcome its application worldwide. We would like to determine whether targeting parameters of the ABLE score can be altered by focused changes in patient care. In conclusion, the CF-ABLE score is a simple, practical clinical prediction tool that can predict prognosis in CF over a 4-year period and has been validated in a national cohort. Acknowledgments Author contributions: Drs McCarthy and Dimitrov had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Dr McCarthy: contributed to the study concept, data analysis and interpretation, preparation of the manuscript, and review and approval of the final manuscript. Dr Dimitrov: contributed to the study design and practical validation of the CF-ABLE score, data analysis and interpretation, and review and approval of the final manuscript. Dr Meurling: contributed to the data interpretation, statistical analysis, and review and approval of the final manuscript. Dr Gunaratnam: contributed to the study concept, data interpretation, preparation of the manuscript, and review and approval of the final manuscript. Dr McElvaney: contributed to the study concept, data interpretation, preparation of the manuscript, and review and approval of the final manuscript. Financial/nonfinancial disclosures: The authors have reported to CHEST that no potential conflicts of interest exist with any companies/organizations whose products or services may be discussed in this article. Additional information: The e-Appendix, e-Figures, and e-Tables can be found in the “Supplemental Materials” area of the online article.
References 1. Ramsey BW. Management of pulmonary disease in patients with cystic fibrosis [published correction in N Engl J Med. 1996;335(15):1167]. N Engl J Med. 1996;335(3):179-188. 2. Corey M, Edwards L, Levison H, Knowles M. Longitudinal analysis of pulmonary function decline in patients with cystic fibrosis. J Pediatr. 1997;131(6):809-814. 3. Rosenbluth DB, Wilson K, Ferkol T, Schuster DP. Lung function decline in cystic fibrosis patients and timing for lung transplantation referral. Chest. 2004;126(2):412-419. 4. Amadori A, Antonelli A, Balteri I, Schreiber A, Bugiani M, De Rose V. Recurrent exacerbations affect FEV(1) decline in adult patients with cystic fibrosis. Respir Med. 2009;103(3): 407-413.
5. Sanders DB, Bittner RC, Rosenfeld M, Redding GJ, Goss CH. Pulmonary exacerbations are associated with subsequent FEV1 decline in both adults and children with cystic fibrosis. Pediatr Pulmonol. 2011;46(4):393-400. 6. Kerem E, Reisman J, Corey M, Canny GJ, Levison H. Prediction of mortality in patients with cystic fibrosis. N Engl J Med. 1992;326(18):1187-1191. 7. VanDevanter DR, Wagener JS, Pasta DJ, et al. Pulmonary outcome prediction (POP) tools for cystic fibrosis patients. Pediatr Pulmonol. 2010;45(12):1156-1166. 8. Liou TG, Adler FR, Fitzsimmons SC, Cahill BC, Hibbs JR, Marshall BC. Predictive 5-year survivorship model of cystic fibrosis. Am J Epidemiol. 2001;153(4):345-352. 9. Taussig LMKJ, Kattwinkel J, Friedewald WT, Di Sant’Agnese PA. A new prognostic score and clinical evaluation system for cystic fibrosis. J Pediatr. 1973;82(3):380-390. 10. Hafen GM, Ranganathan SC, Robertson CF, Robinson PJ. Clinical scoring systems in cystic fibrosis. Pediatr Pulmonol. 2006;41(7):602-617. 11. Fogarty AW, Britton J, Clayton A, Smyth AR. Are measures of body habitus associated with mortality in cystic fibrosis? Chest. 2012;142(3):712-717. 12. Courtney JM, Bradley J, Mccaughan J, et al. Predictors of mortality in adults with cystic fibrosis. Ped Pulmonol. 2007; 42(6):525-532. 13. Milla CE, Warwick WJ. Risk of death in cystic fibrosis patients with severely compromised lung function. Chest. 1998;113(5): 1230-1234. 14. Mayer-Hamblett N, Rosenfeld M, Emerson J, Goss CH, Aitken ML. Developing cystic fibrosis lung transplant referral criteria using predictors of 2-year mortality. Am J Respir Crit Care Med. 2002;166(12):1550-1555. 15. Konstan MW, Morgan WJ, Butler SM, et al; Scientific Advisory Group and the Investigators and Coordinators of the Epidemiologic Study of Cystic Fibrosis. Risk factors for rate of decline in forced expiratory volume in one second in children and adolescents with cystic fibrosis. J Pediatr. 2007; 151(2):134-139. 16. Waters V, Stanojevic S, Atenafu EG, et al. Effect of pulmonary exacerbations on long-term lung function decline in cystic fibrosis. Eur Respir J. 2012;40(1):61-66. 17. de Boer K, Vandemheen KL, Tullis E, et al. Exacerbation frequency and clinical outcomes in adult patients with cystic fibrosis. Thorax. 2011;66(8):680-685. 18. Shwachman H, Kulczycki LL. Long-term study of one hundred five patients with cystic fibrosis; studies made over a five- to fourteen-year period. AMA J Dis Child. 1958;96(1):6-15. 19. Doershuk CF, Matthews LW, Tucker AS, et al. A 5 year clinical evaluation of a therapeutic program for patients with cystic fibrosis. J Pediatr. 1964;65:677-693. 20. Cooperman EM, Park M, McKee J, Assad JP. A simplified cystic fibrosis scoring system (a preliminary report). Can Med Assoc J. 1971;105(6):580-582.
1364
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
Original Research GENETIC AND DEVELOPMENTAL DISORDERS
The CF-ABLE Score A Novel Clinical Prediction Rule for Prognosis in Patients With Cystic Fibrosis Cormac McCarthy, MB; Borislav D. Dimitrov, PhD; Imran J. Meurling, MB; Cedric Gunaratnam, MB; and Noel G. McElvaney, MB
Background: Determining prognosis and predicting outcomes in cystic fibrosis (CF) is a complex issue, and there have been very few clinically applicable models for this. The aim was to create a simple, practical outcome prediction tool for CF. Methods: Forty-nine consecutive patients with CF from a single center were studied over an 84-month period (2004-2010). All baseline clinical parameters were gathered, and FEV1 measurements were analyzed over the study period. Using patterns of FEV1 decline, a tipping point of 52.8% predicted was identified. Other clinical variables were analyzed and correlated with outcome. Poor outcome was defined as death or transplantation. Using age, BMI, lung function (ie, FEV1), and number of exacerbations in the past 3 months, the CF-ABLE score was created. The score was validated for data from 370 patients from the national Cystic Fibrosis Registry of Ireland. Results: The ABLE score uses clinical parameters that are measured at every clinic visit and scored on a scale from 0 to 7. If FEV1 is , 52%, then 3.5 points are added; if the number of exacerbations in the past 3 months is . 1, then 1.5 points are added; if BMI is , 20.1 kg/m2 or age , 24 years, each receive 1 point. Conclusions: Patients with a low score have a very low risk of death or lung transplantation within 4 years; however, as the score increases, the risk significantly increases. Patients who score . 5 points have a 26% risk of poor outcome within 4 years. This score is simple and applicable and better CHEST 2013; 143(5):1358–1364 predicts outcome than FEV1 alone. Abbreviations: CF 5 cystic fibrosis; CFRI 5 Cystic Fibrosis Registry of Ireland; ROC 5 receiver operating characteristic
fibrosis (CF) is a multisystem disorder with a Cystic significantly shortened life expectancy. The major
cause of mortality in CF is lung disease.1 As the condition worsens over time, lung function declines. This decline usually is measured by FEV1.2 Previous studies suggested that patients with CF with an earlier mortality demonstrate a greater rate of decline in FEV1% predicted compared to patients with CF with Manuscript received August 14, 2012; revision accepted October 17, 2012. Affiliations: From the Respiratory Research Division (Drs McCarthy, Meurling, Gunaratnam, and McElvaney), Department of Medicine, and Department of General Practice (Dr Dimitrov), Royal College of Surgeons in Ireland, Dublin, Ireland; Academic Unit of Primary Care and Population Sciences (Dr Dimitrov), University of Southampton, Southampton, England; and Department of Respiratory Medicine (Drs McCarthy, Meurling, Gunaratnam, and McElvaney), Beaumont Hospital, Dublin, Ireland. Funding/Support: The authors have reported to CHEST that no funding was received for this study.
later mortality/longer survival.3 In this regard, decrements in lung function are associated with recurrent pulmonary exacerbations, which are common in CF, and frequent exacerbations lead to an increased rate of decline in FEV1.4,5 Decreased FEV1 at a young age has also been shown to correlate with a poor outcome.6 Many studies investigated the accuracy and usefulness of prediction tools in CF. Both clinical findings and lung function parameters have been used to create prediction models for pulmonary outcomes.7 Most have been hampered by complexity and difficulty in Correspondence to: Cormac McCarthy, MB, Respiratory Research Division, Department of Medicine, Royal College of Surgeons in Ireland, Education and Research Centre, Beaumont Hospital, Beaumont Rd, Dublin 9, Ireland; e-mail: [email protected] © 2013 American College of Chest Physicians. Reproduction of this article is prohibited without written permission from the American College of Chest Physicians. See online for more details. DOI: 10.1378/chest.12-2022
1358
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
application.8,9 This complexity has precluded these prediction tools from being used routinely, and hence, many clinical decisions are guided by FEV1 alone. A better and more practical prediction tool is needed to stratify risk and predict outcome earlier to facilitate interventions or to guide transplant assessment at an earlier stage.10 The aim of the present study was to create an easy-to-use scoring tool that predicts risk over a significant time period using commonly measured clinical parameters. The ideal prediction tool should be one that is simple and quick to score and uses relevant clinical information that can be addressed to improve outcome if necessary. Because BMI has been shown to correlate with increased mortality risk, it was also assessed in the prediction algorithm.11,12 The prediction tool was validated in a larger cohort from the national Cystic Fibrosis Registry of Ireland (CFRI). Materials and Methods Study Design and Selection of Patients This was a prospective, single-cohort study for the derivation and validation of a new clinical prediction rule, the CF-ABLE score. Approval for this study was obtained from the ethics committee of Beaumont Hospital (approval number 12/80). Initial data were obtained from a single specialist adult CF center. A cohort group of 49 consecutive patients were included in the derivation part of the study. Data were collected on pancreatic insufficiency, diabetes mellitus status, Pseudomonas aeruginosa colonization, Staphylococcus aureus colonization, genotype, FEV1, BMI, bone mineral density, and number of exacerbations. Patient FEV1 measurements from 2004 to 2009 were obtained from medical records, and all data on pulmonary exacerbations for this period were evaluated at the end of 2009. Exacerbations were defined as events in which the patient’s clinical condition changed significantly enough to require treatment with IV antibiotics administered either in the hospital or in the home setting. BMI was recorded, and initial BMI was the first measurement noted in the period. Patient age was recorded at the time of the first measurements made in the study period. At the end of 2010, the patients were retrospectively classified according to outcome, which was defined as good or poor outcome. Patients were classified with poor outcome if they were referred for lung transplantation or had died in the interim. Criteria for lung transplantation referral were a baseline FEV1 , 30% predicted, oxygen saturation , 90% at rest or on exertion, or both. Following identification of patterns of lung function decline and predictors of poor outcome in the initial cohort and derivation of the CF-ABLE rule, new data from the national CFRI were collected for validation purposes. We gathered the data for all available patients from the entire registry from 2005 to 2010 with the exception of the 49 already analyzed in the derivation cohort. Patients were excluded if they were aged , 16 years, if they had already received a double lung transplantation prior to 2005, or if no pulmonary function or BMI data were available prior to their transplant. Statistical Analysis Because this was a pilot study, a sample size was not calculated a priori; however, a post hoc calculation of power was performed,
as appropriate. Data are presented as mean ⫾ SD for continuous variables or number and frequency (percentage, proportion) for categorical variables and analyzed with two-tailed parametric (t) or nonparametric (x2, Mann-Whitney) tests, as appropriate. The analyses at group (population) level used parametric regression modeling with fitting of a number of linear and nonlinear equations to analyze and project the group FEV1 percent predicted by a range of built-in linear, cubic, polynomial, and other high-order functions (Fig 1). Logistic regression analysis was applied by entry and backward stepwise methods, with adjustment for covariate effects for variables significantly associated with poor prognosis in the univariable analyses or found clinically relevant and, eventually, important. Survival curves were based on Kaplan-Meier estimates. To test the independence over time of the predicting variables from the best logistic regression models, the cumulative hazard function of the primary end point was calculated by means of Cox regression modeling, with adjustment for covariate effects. To confirm the predictive performance of the best model, we further applied a receiving operator characteristic (ROC) curve analysis on the probabilities derived from the best model, identifying the best cutoff point for each univariate, statistically significant predictor. The statistical significance of all tests was assumed at P , .05, unless stated otherwise. All evaluations were done with SPSS version 18 (International Business Machines Corporation) statistical software. Initially, to explore various hypotheses about possible predictors of CF prognosis, we collected all pulmonary function data for patients over a 6-year period (2004-2009), which were transformed into single values for equidistant 3-month periods as means of all valid measurements for each patient. Pulmonary exacerbations were calculated as a total number for the period and subcategorized as number of exacerbations per 3-month period. All demographic and clinical variables are described and presented by their main statistical parameters, overall, and by the type of outcome. Algorithm for Derivation of CF-ABLE Score Univariate and multifactorial logistic regression analyses were applied to identify and select the significant independent predictors. Univariate analyses and ROC curves were used to identify the best cutoff points in terms of individual outcome for each predictor (e-Fig 1). Predictors were entered into a multifactor regression model with other potential predictors. The predictive performance of the best model and its individual predictors were tested by ROC curve analysis, where best cutoff points were established for each predictor according to the results and clinical judgment. The best cutoff levels were used to create new dichotomized or trichotomized variables from each single or combined independent predictor to specify its initial scores or high-risk and low-risk levels (ie, presence [1] or absence [0] of risk). Values were dichotomized or trichotomized to create a simple clinical prediction score. Subsequently, a new logistic regression model with these dichotomized/trichotomized values was constructed (e-Appendix 1, section I). The distribution of outcome according to the different levels of the CF-ABLE score as well as its predictive performance by ROC curve analysis was also computed (e-Fig 2). After the construction of the CF-ABLE score in the derivation cohort, the same parameters were taken from the national data set, and the score was further validated in the CFRI.
Results The total number of patients from the initial center was 49; 36 were classified as having had a good outcome at the end of the study in 2010 (ie, none of
journal.publications.chestnet.org
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
CHEST / 143 / 5 / MAY 2013
1359
Figure 1. Model of lung function decline in the derivation cohort. A, Demonstrates mean FEV1 percent predicted for the entire cohort plotted against time. All FEV1 measurements were transformed into single values of means for equidistant 3-month periods. These are plotted against time to demonstrate the overall pattern of decline in lung function in the derivation cohort as a group. B, Parametric regression modeling was used to analyze the pattern of FEV1 decline at the group level. Fitting of a number of linear and nonlinear equations to analyze and project the decline was done by a range of built-in linear, cubic, polynomial, and other high-order functions. There was a significant nonlinear tendency of decline in mean FEV1. The most statistically significant models of decline were a cubic equation (red curve) (P , .05), showing an apparent plateau phase and a cyclic pattern. Additionally, an inverse equation (blue curve) (P , .05) showed a rapid decline followed by a prolonged plateau phase.
these patients died or were referred for lung transplantation). They had an average initial FEV1 of 82.91% and a mean number of 1.27 exacerbations per 3-month period. Thirteen patients were classified as having had a poor outcome, of whom 10 were referred for lung transplantation (two died before 2010) and two
others received lung transplantation in 2010. Three other patients died before 2010, none of whom were referred for lung transplantation. The poor-outcome group had a lower average initial FEV1 of 39.58% and a higher mean number of exacerbations per 3-month period of 1.54 (Table 1).
1360
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
Table 1—Demographics of the Derivation and Validation Cohorts Clinical and Demographic Parameter Derivation Cohort No. patients Patients who died Patients who received or were referred for transplant Patients with poor outcome (ie, died/ transplantation) Age at first FEV1, y Male sex Female sex DF508 homozygous Initial BMI, kg/m2 Initial FEV1, % predicted No. exacerbations in initial 3 months Range
Validation Cohort
49 5 (10.2) 10 (20.4)
370 23 (6.2) 11 (2.9)
13 (26.5)
34 (9)
21.24 (20.13-22.35) 24.27 (23.55-24.99) 31 (63.27) 211 (57) 18 (36.73) 159 (43) 27 (55) 212 (57.3) 20.41 (19.39-21.47) 21.38 (20.98-21.78) 62.05 (54.76-69.34) 60.16 (57.62-62.7) 0.36 (0.12-0.54) 0.34 (0.281-0.398) 0-4
0-3
Data are presented as No. (%) or mean (95% CI). DF508 5 mutation of the CFTR gene.
The analysis of all FEV1 measurements over the 6-year period showed a significant, nonlinear tendency of decline in mean FEV1 over time (P , .05) (Fig 1A) that was indicative of a tipping point. The mean estimates from a generalized linear model were best described by both a cubic equation (P , .05) (Fig 1B) showing an apparent plateau phase in lung function (with a possible cyclic pattern) and an inverse equation (P , .05) (Fig 1B) showing a rapid decline followed by a prolonged plateau phase in FEV1 decline. The modeling also identified a parallel, nonlinear increase in the mean number of exacerbations (per 3-month period) over time (P , .05) that correlated significantly with the decline in lung function (P , .05) for all patients. The poor-outcome group had a significantly greater number of exacerbations throughout the initial 4-year period (2004-2007) compared with the good-outcome group (P 5 .02) and had a significantly greater mean number of exacerbations in the initial 3-month period. The poor-outcome group had a mean number of exacerbations of two per 3-month period initially and an average of 1.54 per 3-month period over the 4 years; the mean exacerbation rate in the good-outcome group was 1.27 per 3-month (Fig 2). The mean initial BMI of the poor-outcome group was 19.6 kg/m2, and the mean age was 21.89 years; BMI and age of the goodoutcome group were 20.63 kg/m2 and 21.01 years. Given the results from the analyses at the group level (Fig 1B, e-Figs 3, 4), we used the regression analysis and ROC curve approach to identify a value around the average FEV1 of 56.75% predicted for the good-outcome group, which served as the cutoff point to distinguish patients in this group from those
Figure 2. Mean number of exacerbations per 3-month period in the derivation cohort. The mean number of exacerbations per 3-month interval in good outcome and poor outcome throughout the initial 4-year period (2004-2007) are shown. The pooroutcome group had a greater number of exacerbations and more clustering.
in the poor-outcome group. Through a ROC curve analysis (e-Fig 1), we identified an FEV1 cutoff of 52% to 53% for initial FEV1 measurements and a similar value for mean FEV1 in the second quarter (52.8%). Patients with an FEV1 above this level had a significantly lower risk of worse outcome in the following 4 years. Derivation of the CF-ABLE Score The main univariate analysis and ROC curve identified a best cutoff point in terms of individual outcome (e-Fig 1). This main predictor (mean FEV1% predicted during the first 3 months) was then entered into a multifactor regression model with other potential predictors (BMI, age at first FEV1% predicted, mean number of exacerbations during first 3 months) (e-Appendix 1, sections I, II, III, e-Table 1) The predictive performance of the best model and its individual predictors were tested by ROC curve analysis, where best cutoff points were established for each predictor according to the results and clinical judgment in order to incorporate the four variables as predictive components of a new, simplified score. The new rule to predict poor outcome was named the ABLE score according to its components (age, BMI, lung function [ie, FEV1% predicted], and exacerbations) and constructed by the algorithm described next. The best cutoff levels were used to create new dichotomized or trichotomized variables from each single or combined independent predictor in order to specify its initial scores or high-risk and low-risk levels (ie, presence [1] or absence [0] of risk ) (e-Appendix 1, section I). The next step was to create a new logistic regression model with the dichotomized and trichotomized values (e-Appendix 1, section I). Using this methodology, the ABLE score is constructed from the sum
journal.publications.chestnet.org
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
CHEST / 143 / 5 / MAY 2013
1361
of the individual component scores and has the following ranges: ABLE score [0.0 2 7.0] 5 FEV1% predicted (mean previous 3 months) [0.0 2 3.5] 1 number exacerbations (previous 3 months) [0.0 2 1.5] 1 age at time of assessment 3 BMI [0.0 2 2.0] Validation of the CF-ABLE Score in the National CFRI Cohort During the period of 2005 to 2010, data from 370 patients from the national CFRI were included in the analysis. Twenty-three patients died and 11 patients underwent double lung transplantation for an event rate of 9.19% for the poor-outcome group (Table 1). The application of the new clinical prediction rule in the national CFRI validation cohort ranged from 0 to 7 (2.84 ⫾ 2.22). There was a significant difference between the mean ABLE scores of the two subgroups by outcome (good outcome, 2.61 ⫾ 2.15; poor outcome, 5.21 ⫾ 1.30; P , .001). The histograms of the distributions were also computed and clearly illustrated the statistically significant difference (Fig 3). The discriminative performance was further confirmed by a new ROC curve analysis (e-Fig 5), indicating a significant area under the ROC curve of 0.82 (95% CI, 0.77-0.88). The logistic regression model to predict
death or referral for transplantation in the validation cohort indicated that, on average, a 1-point increase of ABLE score relative to the previous score would lead to an 81% increase in risk (OR, 1.81; 95% CI, 1.46-2.22); for example, an increase in ABLE score from 3.50 to 4.50 would lead to an increased risk of 6.45% to 14.71%. In terms of calibration performance, it can be seen that with the increase of ABLE score, the risk increases linearly along the whole range of the score (Fig 4A, e-Appendix 1, section IV; e-Table 2, 3; e-Fig 6). Discussion We developed an easy-to-use predictive score for determining outcome in CF. This score accurately predicts the risk of death or transplantation within 4 years, as validated in the national CFRI. Previous studies identified an FEV1 , 30% predicted to be a predictor of mortality and have suggested transplant referral at this point.6 Others argued that this value is not predictive of outcome and that rate of lung function decline is more sensitive.13 In the present study, we demonstrate that there is a nonlinear decline in FEV1% predicted over time, which follows one of two patterns: either a sharp decline followed by plateau that declines further in a cyclical manner or a sharp decline followed by a prolonged
Figure 3. Histograms of distribution of the cystic fibrosis (CF)-ABLE score within both outcome groups in the validation cohort. Application of the ABLE score in the validation cohort demonstrated a statistically significant difference in the mean ABLE scores between the good- and poor-outcome groups (P , .001). The mean ⫾ SD CF-ABLE score in the good-outcome group was 2.61 ⫾ 2.15 and in the poor-outcome group, 5.21 ⫾ 1.3. 1362
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research
Figure 4. A, Risk of poor outcome according to CF-ABLE score in the validation cohort. There is a linear correlation in risk for a poor outcome (death or transplantation) within 4 years as ABLE score increases. In the validation cohort, there was a significant change in risk as the ABLE score increased to . 3.5. There is a 26% chance of poor outcome within 4 years if the ABLE score is ⱖ 5. B, CF-ABLE scoring table. The ABLE score is based on adding the values for each variable of the score. The total is from 0 to 7. As the score increases, there is a linear increase in risk of poor outcome. See Figure 3 legend for expansion of abbreviation.
plateau phase.14 We identify that an FEV1 of 52% predicted is an accurate predictor of outcome and identifies high-risk patients earlier. We demonstrate that in the poor-outcome group, there was a significantly higher number of exacerbations per 3-month period over the course of the study and that the clustering of exacerbations has a deleterious effect on long-term lung function measurements and survival.4,5,15-17 Thus, the number of exacerbations is another key component of our predictive model. Finally, we identify age as having a significant effect on risk prediction, especially in patients with a low FEV1 or low BMI at a young age. This further underscores the need for goal-driven nutritional strategies to improve survival in patients with CF. Several prediction models and scoring tools have been designed for use in CF; however, most have been either too complex to score, cumbersome to use, or unreliable. For example, the Shwachman-Kulczycki score relies heavily on subjective assessments, requires a combination of radiographic and clinical parameters, is difficult to score, and is not acceptably reliable.18,19 Similarly, the Cooperman and National Institutes of Health scores are subjective and overly complex and have not been validated.9,20 There are few validated prediction rules in CF. Liou et al8 created a survivorship
model from US CF registry data. This is an efficient prediction model and has been validated in a large cohort of registry patients, but it is somewhat limited by deriving the prediction rule from the same cohort as the validation group. Additionally, the scoring system with eight parameters and at least 15 steps in the computation make it complex and difficult to use. In comparison, the CF-ABLE score has only four parameters and is applied by using very simple rules and a straightforward arithmetic calculation (Fig 4B). Furthermore, the score by Liou et al8 has been validated only by Hosmer-Lemeshow test, and no ROC analyses or any other accuracy measures were shown, unlike with the CF-ABLE score. There are several potential limitations to the present study. We had a small derivation population, but when compared with the national CFRI the patients had similar outcomes and demographics. The CF-ABLE scores in the derivation and validation cohorts were very comparable, and regarding the slight differences noted between the groups, there are a number of reasons for this. First, a higher number of events were captured in the small cohort of patients from the derivation group because this group was closely evaluated over the period of study. Second, referral for transplant was used as a proxy measure for transplantation in the derivation cohort as a clinical outcome. Third, the data collected for the validation cohort were limited by the number of patients in the registry who were not included in the study because they had undergone transplantation but had no data available in the registry prior to their transplantation. Finally, a potential problem with this scoring system is the difficulty in defining a pulmonary exacerbation in CF. We used IV antibiotics treatment at home or in the hospital as definition for a significant acute exacerbation, which may have underestimated the total number of exacerbations; however, we are confident that we identified all the severe exacerbations. This study adds to the field of stratifying patient risks at earlier time points to identify the need for earlier intervention with new treatments or alternative ways of delivering the best available treatment. It may direct CF clinicians to concentrate their therapeutic efforts on aspects of care most amenable to change. The CF-ABLE score identifies both highand low-risk patients over a prolonged period. It demonstrates that those with a low score (eg, , 2) have a small risk of death or transplantation in the next 4 years, and this should encourage clinicians and their patients to maintain care and CF-ABLE score parameters. If the patient has a high score (eg, 5 of 7), he or she has a 26% risk of death or need for a lung transplantation in the next 4 years. We believe that this tool is a robust and effective means to assess risk in CF. It has been validated in a
journal.publications.chestnet.org
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
CHEST / 143 / 5 / MAY 2013
1363
large CF population with the common CF mutations. As such, we believe that it is generalized beyond Ireland, and we would welcome its application worldwide. We would like to determine whether targeting parameters of the ABLE score can be altered by focused changes in patient care. In conclusion, the CF-ABLE score is a simple, practical clinical prediction tool that can predict prognosis in CF over a 4-year period and has been validated in a national cohort. Acknowledgments Author contributions: Drs McCarthy and Dimitrov had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Dr McCarthy: contributed to the study concept, data analysis and interpretation, preparation of the manuscript, and review and approval of the final manuscript. Dr Dimitrov: contributed to the study design and practical validation of the CF-ABLE score, data analysis and interpretation, and review and approval of the final manuscript. Dr Meurling: contributed to the data interpretation, statistical analysis, and review and approval of the final manuscript. Dr Gunaratnam: contributed to the study concept, data interpretation, preparation of the manuscript, and review and approval of the final manuscript. Dr McElvaney: contributed to the study concept, data interpretation, preparation of the manuscript, and review and approval of the final manuscript. Financial/nonfinancial disclosures: The authors have reported to CHEST that no potential conflicts of interest exist with any companies/organizations whose products or services may be discussed in this article. Additional information: The e-Appendix, e-Figures, and e-Tables can be found in the “Supplemental Materials” area of the online article.
References 1. Ramsey BW. Management of pulmonary disease in patients with cystic fibrosis [published correction in N Engl J Med. 1996;335(15):1167]. N Engl J Med. 1996;335(3):179-188. 2. Corey M, Edwards L, Levison H, Knowles M. Longitudinal analysis of pulmonary function decline in patients with cystic fibrosis. J Pediatr. 1997;131(6):809-814. 3. Rosenbluth DB, Wilson K, Ferkol T, Schuster DP. Lung function decline in cystic fibrosis patients and timing for lung transplantation referral. Chest. 2004;126(2):412-419. 4. Amadori A, Antonelli A, Balteri I, Schreiber A, Bugiani M, De Rose V. Recurrent exacerbations affect FEV(1) decline in adult patients with cystic fibrosis. Respir Med. 2009;103(3): 407-413.
5. Sanders DB, Bittner RC, Rosenfeld M, Redding GJ, Goss CH. Pulmonary exacerbations are associated with subsequent FEV1 decline in both adults and children with cystic fibrosis. Pediatr Pulmonol. 2011;46(4):393-400. 6. Kerem E, Reisman J, Corey M, Canny GJ, Levison H. Prediction of mortality in patients with cystic fibrosis. N Engl J Med. 1992;326(18):1187-1191. 7. VanDevanter DR, Wagener JS, Pasta DJ, et al. Pulmonary outcome prediction (POP) tools for cystic fibrosis patients. Pediatr Pulmonol. 2010;45(12):1156-1166. 8. Liou TG, Adler FR, Fitzsimmons SC, Cahill BC, Hibbs JR, Marshall BC. Predictive 5-year survivorship model of cystic fibrosis. Am J Epidemiol. 2001;153(4):345-352. 9. Taussig LMKJ, Kattwinkel J, Friedewald WT, Di Sant’Agnese PA. A new prognostic score and clinical evaluation system for cystic fibrosis. J Pediatr. 1973;82(3):380-390. 10. Hafen GM, Ranganathan SC, Robertson CF, Robinson PJ. Clinical scoring systems in cystic fibrosis. Pediatr Pulmonol. 2006;41(7):602-617. 11. Fogarty AW, Britton J, Clayton A, Smyth AR. Are measures of body habitus associated with mortality in cystic fibrosis? Chest. 2012;142(3):712-717. 12. Courtney JM, Bradley J, Mccaughan J, et al. Predictors of mortality in adults with cystic fibrosis. Ped Pulmonol. 2007; 42(6):525-532. 13. Milla CE, Warwick WJ. Risk of death in cystic fibrosis patients with severely compromised lung function. Chest. 1998;113(5): 1230-1234. 14. Mayer-Hamblett N, Rosenfeld M, Emerson J, Goss CH, Aitken ML. Developing cystic fibrosis lung transplant referral criteria using predictors of 2-year mortality. Am J Respir Crit Care Med. 2002;166(12):1550-1555. 15. Konstan MW, Morgan WJ, Butler SM, et al; Scientific Advisory Group and the Investigators and Coordinators of the Epidemiologic Study of Cystic Fibrosis. Risk factors for rate of decline in forced expiratory volume in one second in children and adolescents with cystic fibrosis. J Pediatr. 2007; 151(2):134-139. 16. Waters V, Stanojevic S, Atenafu EG, et al. Effect of pulmonary exacerbations on long-term lung function decline in cystic fibrosis. Eur Respir J. 2012;40(1):61-66. 17. de Boer K, Vandemheen KL, Tullis E, et al. Exacerbation frequency and clinical outcomes in adult patients with cystic fibrosis. Thorax. 2011;66(8):680-685. 18. Shwachman H, Kulczycki LL. Long-term study of one hundred five patients with cystic fibrosis; studies made over a five- to fourteen-year period. AMA J Dis Child. 1958;96(1):6-15. 19. Doershuk CF, Matthews LW, Tucker AS, et al. A 5 year clinical evaluation of a therapeutic program for patients with cystic fibrosis. J Pediatr. 1964;65:677-693. 20. Cooperman EM, Park M, McKee J, Assad JP. A simplified cystic fibrosis scoring system (a preliminary report). Can Med Assoc J. 1971;105(6):580-582.
1364
Downloaded From: http://journal.publications.chestnet.org/ by David Kinnison on 05/15/2013
Original Research