0
Communications to the Editor |

Testing Lung Function Decline to Time Lung Transplantation FREE TO VIEW

Theodore G. Liou, MD, FCCP; Frederick R. Adler, PhD; Barbara C. Cahill, MD
Author and Funding Information

Affiliations: Department of Internal Medicine,  Departments of Mathematics and Biology,  Department of Internal Medicine, University of Utah, Salt Lake City, UT,  Washington University School of Medicine, St. Louis, MO

Correspondence to: Theodore G. Liou, MD, Division of Respiratory, Critical Care and Occupational Pulmonary Medicine, 26 North Medical Dr, Salt Lake City, UT 84132; e-mail: ted.liou@utah.edu.



Chest. 2005;128(1):472-474. doi:10.1378/chest.128.1.472
Text Size: A A A
Published online

To the Editor:

Timing of referral for lung transplantation for patients with end-stage lung disease from cystic fibrosis (CF) remains a difficult issue. The recent publication by Rosenbluth et al1 proposes a method for timing referral of patients for wait listing. Using at least 4 years of previously measured percentage of predicted FEV1 (FEV1%), Rosenbluth et al suggest using a linear regression model to predict when FEV1% drops to < 30% of predicted and suggest listing patients for transplantation 2.11 years prior to that point in order to minimize deaths due to late referrals for lung transplantation.

We have two serious concerns regarding this work. First, the linear regression model proposed is not validated and may lack predictive power. Second, even were this model predictive, elimination of late referrals for lung transplantation will not change the inability of the FEV1% criterion by itself to identify patients who will have survival benefit from transplantation.

To examine these issues, we undertook validation of the proposed model and projected the outcomes for patients likely to receive referral for transplantation using the Rosenbluth model. We used the CF Foundation Patient Registry (CFFPR) to identify the 5,408 patients in 1999 who had at least 4 prior consecutive years of FEV1% data, did not undergo transplantation prior to December 31, 2001, and had FEV1% data in 2001. We used linear regression (based on the highest measured FEV1% in a given calendar year) to predict the FEV1% in 2001. Using additional data from the CFFPR, we calculated the 5-year predicted survival2 of each patient for 2001 to see how well the proposed model chooses patients for lung transplantation. Figure 1 compares predicted change in FEV1% with actual change in FEV1% between 1999 and 2001. Among the patients shown in Figure 1, there were 458 patients that were predicted to have an FEV1% < 30% in 2001. Figure 2 shows the actual FEV1% in 2001 for those 458 patients compared to their 5-year predicted survivals in 2001.

These results allow us to address our two concerns. First, the method should be accurate in predicting that FEV1% will be < 30% in 2001. Unfortunately, there was no correlation between predicted and actual changes in FEV1% (Fig 1). For patients predicted to have an FEV1% < 30%, 45% of the actual FEV1% values in 2001 were > 30% (Fig 2). As we have shown previously, this is due to the low predictive power of linear regression to predict future FEV1%.3

Second, of all patients identified by Rosenbluth et al1 as good candidates for lung transplantation using FEV1% < 30%, only 15% would have predicted improved survival, while 22% would have predicted unchanged survival, and 63% would have predicted reduced survival based on their 5-year predicted survivals (Fig 2).,4 In fact, even if we restrict attention to those 55% of patients who actually have FEV1% < 30% in 2001, fully half of those patients selected are not appropriate candidates for transplantation. We base these conclusions on our previous finding that using FEV1% by itself cannot identify a group of patients that has survival benefits from transplantation.,4 In contrast, our 5-year predicted survival model does identify a group that triples its survival due to transplantation (Liou et al,4 Figure panels A and F).

Rosenbluth et al1 have cited our work4 as justification for using FEV1% < 30% as a criterion for lung transplantation because those who had benefit from the procedure also had a mean FEV1% < 30%. Although the group that we chose for transplantation in our work had an FEV1% of 23% (calculated from Table 2 of Liou et al,4), fully 72% of the patients with FEV1% < 30% were not appropriate candidates for transplantation. Patients deemed inappropriate despite their low FEV1% actually had a lower mean FEV1% (21%, calculated from the original data set,4) than those patients who had benefit from transplantation (t test, p = 0.0017). The latter group of patients is clinically indistinguishable based on FEV1% alone, but derives no survival benefit from transplantation. Because of other clinically relevant factors such as microbiology and nutrition, these patients survive longer, on average, without transplantation.

The authors discuss using rate of decline of FEV1% to stratify patients with varying severities of CF, yet as we show here the rate of decline method that they propose does not validate. This reproduces our original finding (but with a new data set) that rate of decline of FEV1% derived either by linear regression or mixed effects modeling has no predictive value for survival in CF using CFFPR data from 1986 to 1993.23

Our validation of the Rosenbluth et al1 patient selection method has one potential weakness. Although the CFFPR contains information on > 30,000 patients with CF and is estimated to include ≥ 85% of all such patients in the United States,5 the specific pulmonary measures for each patient have been limited to no more than four per year. For many patients, there is only one measure in a given year, and for most patients the date of pulmonary testing is not precise. Centers are instructed to submit the best pulmonary measures for each quarter, but we are unable to verify compliance. These issues may have introduced enough noise to reduce the predictive power of linear regression using the CFFPR. Nevertheless, this potential weakness does not affect the inability of the FEV1% < 30% criterion to select appropriate candidates for lung transplantation.

In addition to the failure to validate or to choose the right patients for transplantation, the method proposed by Rosenbluth et al1 would be quite complex to implement in practice. First, it requires multiple years of high-quality data. Second, it requires choosing appropriate time windows over which to evaluate the decline in FEV1%. Rosenbluth et al, in their Figure 2 and accompanying text,,1 present no objective method for selecting the correct data points for application of linear regression. Finally, for transplant referral centers that do not have direct access to a patient’s past data, it increases the difficulty to corroborate the transplant decision. In contrast, while our 5-year predicted survival model requires multiple variables, they are all obtained as part of routine CF care and can be collected during a single clinic visit. A work sheet is published that makes calculation of 5-year predicted survival straightforward (aje.oupjournals.org/cgi/content/full/153/4/345/DC1).

The method proposed by Rosenbluth et al1for selecting patients for lung transplantation is not simple and will lead to no survival benefit. However, there is a validated method2 that identified patients for whom lung transplantation clearly increased survival.4 Although rate of decline of lung function may prove useful in the future to further improve patient selection, clinical variables with a current demonstrated ability to improve selection should be utilized first.

Figure Jump LinkFigure 1. The actual changes in FEV1% between 1999 and 2001 for 5,408 patients are compared to the predicted changes in FEV1%. Predictions were derived by linear regression from FEV1% values from 1996 through 1999. The dashed line shows the expected relationship between predicted and actual changes.Grahic Jump Location
Figure Jump LinkFigure 2. Patients predicted to have an FEV1% < 30% in 2001 are shown. The 5-year predicted survival is compared to the actual FEV1% in 2001, the projected time of transplantation. The vertical dashed line separates patients into those above and below an FEV1% of 30%. Nearly half of those predicted to have an FEV1% < 30% have an FEV1% > 30%. The horizontal dashed lines separate the patients into 5-year predicted survival groups: 0 to 30%, 30 to 50% and > 50%. Lung transplantation is expected to increase survival in the low predicted survival group, produce unchanged survival in the middle group, and reduce survival in the high group.Grahic Jump Location
Rosenbluth, DB, Wilson, K, Ferkol, T, et al (2004) Lung function decline in cystic fibrosis patients and timing for lung transplantation.Chest126,412-419. [CrossRef] [PubMed]
 
Liou, TG, Adler, FR, Fitzsimmons, SC, et al Predictive five year survivorship model of cystic fibrosis.Am J Epidemiol2001;153,345-352. [CrossRef] [PubMed]
 
Liou, TG, Adler, FR, Cahill, BC, et al Priorities for lung transplantation among patients with cystic fibrosis.JAMA2002;287,1523-1524. [CrossRef] [PubMed]
 
Liou, TG, Adler, FR, Cahill, BC, et al Survival effect of lung transplantation for patients with cystic fibrosis.JAMA2001;286,2686-2689
 
 Cystic Fibrosis Foundation: Cystic Fibrosis Foundation Patient Registry 2002 annual data report to the center directors. 2003; Cystic Fibrosis Foundation. Bethesda, MD:.
 
To the Editor:

Liou et al raised the following two issues regarding our article (August 2004)1: (1) the linear regression model was not validated; and (2) eliminating late referrals will not help identify patients who could benefit from lung transplantation.

We agree that the model we proposed was not validated, in the sense that it was not applied to a separate patient population or used to predict outcome and then compared to an actual outcome in our own patient population. Such studies would require access to data outside of our center or the performance of a longitudinal follow-up study. We would agree that the results should be replicated in a larger patient population from multiple centers before they are adopted as being definitely valid, as is true for all clinical studies from a single center.

Liou et al then go on to present data that seemingly respond to this need by interrogating the Cystic Fibrosis Foundation Patient Registry and conclude that our model was useless in predicting actual changes in FEV1 percent predicted (Fig 1 in their letter). We are reluctant to engage in a dispute about the meaning of their results because in a letter to the editor too little information is available about the methods employed. For instance, legitimate concerns about data quality are raised by Liou et al themselves in their letter. Furthermore, at the very least, it would be important to exactly replicate the methods used in our study before concluding that there were or were not meaningful differences. We have real concerns that Liou et al did not do this. For instance, in our study, 54% of the patients were in the “slow group” showing pulmonary function decline (ie, the rate of decline for each patient was less than the group mean + 2 SEs). The coefficient of determination for the linear regressions in this group averaged only 0.44 ± 0.30. Clearly, linear regressions in such patients cannot predict future events (in this case, for instance, the FEV1 percent predicted for the next year) with great confidence. Of course, by definition, these patients show little change in their pulmonary function, and therefore, with all other things being equal, they would not be likely candidates for lung transplantation referral. We never implied that such patients do not require close clinical observation and follow-up. If circumstances change, the clinician must adapt accordingly.

In contrast, 35% of the patients were in the “rapidly declining” group. In these patients, the mean coefficient of determination was a startling 0.9 ± 0.1. These results would suggest that for these patients the model should be highly predictive. Our conclusion regarding the referral for lung transplantation for these patients is that with this knowledge referrals could have occurred at an earlier time, potentially saving lives.

It is no surprise then that when Liou et al simply lump all the data in the Cystic Fibrosis Foundation Registry together (Fig 1 in their letter) the strategy that we proposed appears to be of little use; but we discount the value of such an analysis. This concern is relevant to both of the issues raised by Liou et al. Nevertheless, we urge Liou et al to submit their analyses of the complete registry database for publication so that their methods and analytic strategies can be properly peer-reviewed.

Liou et al also have mischaracterized and inappropriately applied our regression model to their data set. The regression model is not used to “predict when the FEV1% drops below 30% of predicted” as they state, but, as stated in our article,1 “the rate of decline was used to predict the age at which %FEV1 would reach 20%.” We agree that the decision to proceed with transplantation should not be based solely on a patient’s FEV1 percent predicted and only assert in our article that this model may be utilized in timing the referral for lung transplantation.

Finally, we dispute the contention that our proposed method is not simple. It is of course trivially simple to analyze the four best points (per annum) of FEV1 percent predicted data by linear regression! Whether, for example, the data have been obtained by accepted standards or the referral center would have access to the required prior data are issues that can be addressed in a straightforward manner if and when others corroborate our proposed approach in future studies.

References
Rosenbluth, DB, Wilson, K, Ferkol, T, et al Lung function decline in cystic fibrosis patients and timing for lung transplantation referral.Chest2004;126,412-419. [CrossRef] [PubMed]
 

Figures

Figure Jump LinkFigure 1. The actual changes in FEV1% between 1999 and 2001 for 5,408 patients are compared to the predicted changes in FEV1%. Predictions were derived by linear regression from FEV1% values from 1996 through 1999. The dashed line shows the expected relationship between predicted and actual changes.Grahic Jump Location
Figure Jump LinkFigure 2. Patients predicted to have an FEV1% < 30% in 2001 are shown. The 5-year predicted survival is compared to the actual FEV1% in 2001, the projected time of transplantation. The vertical dashed line separates patients into those above and below an FEV1% of 30%. Nearly half of those predicted to have an FEV1% < 30% have an FEV1% > 30%. The horizontal dashed lines separate the patients into 5-year predicted survival groups: 0 to 30%, 30 to 50% and > 50%. Lung transplantation is expected to increase survival in the low predicted survival group, produce unchanged survival in the middle group, and reduce survival in the high group.Grahic Jump Location

Tables

References

Rosenbluth, DB, Wilson, K, Ferkol, T, et al (2004) Lung function decline in cystic fibrosis patients and timing for lung transplantation.Chest126,412-419. [CrossRef] [PubMed]
 
Liou, TG, Adler, FR, Fitzsimmons, SC, et al Predictive five year survivorship model of cystic fibrosis.Am J Epidemiol2001;153,345-352. [CrossRef] [PubMed]
 
Liou, TG, Adler, FR, Cahill, BC, et al Priorities for lung transplantation among patients with cystic fibrosis.JAMA2002;287,1523-1524. [CrossRef] [PubMed]
 
Liou, TG, Adler, FR, Cahill, BC, et al Survival effect of lung transplantation for patients with cystic fibrosis.JAMA2001;286,2686-2689
 
 Cystic Fibrosis Foundation: Cystic Fibrosis Foundation Patient Registry 2002 annual data report to the center directors. 2003; Cystic Fibrosis Foundation. Bethesda, MD:.
 
Rosenbluth, DB, Wilson, K, Ferkol, T, et al Lung function decline in cystic fibrosis patients and timing for lung transplantation referral.Chest2004;126,412-419. [CrossRef] [PubMed]
 
NOTE:
Citing articles are presented as examples only. In non-demo SCM6 implementation, integration with CrossRef’s "Cited By" API will populate this tab (http://www.crossref.org/citedby.html).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

CHEST Journal Articles
PubMed Articles
  • CHEST Journal
    Print ISSN: 0012-3692
    Online ISSN: 1931-3543