Patient Selection for Lung Volume Reduction Surgery

Patient Selection for Lung Volume Reduction Surgery* An Objective Model Based on Prior Clinical Decisions and Quantitative CT Analysis David S. Gierada, MD; Roger D. Yusen, MD; Ian A. Villanueva, BS; Thomas K. Pilgram, PhD; Richard M. Slone, MD, FCCP; Stephen S. Lefrak, MD, FCCP; and Joel D. Cooper, MD, FCCP

Objectives: We used whole-lung quantitative CT analysis (QCT)—an objective method of evaluating emphysema severity and distribution based on measurement of lung density—to determine whether subjective selection criteria for lung volume reduction surgery are applied consistently and to model the patient selection process, and assessed the relationship of the model to postoperative outcome. Design: Logistic regression analysis using QCT indexes of emphysema and preoperative physiologic test results as the independent variables, and the decision to operate as the dependent variable. Setting: University hospital. Patients: Seventy patients selected for bilateral lung volume reduction surgery and 32 otherwise operable patients excluded from surgery based on subjective assessment of emphysema morphology on chest radiography, CT, and perfusion scintigraphy. Intervention: Bilateral lung volume reduction surgery in the selected group. Measurements and results: Emphysema in patients selected for surgery was more severe overall and in the upper lungs by multiple QCT indexes (p < 0.01, unpaired two-tailed t test). Physiologic abnormalities were slightly more severe in selected patients (p < 0.05, unpaired two-tailed t test). The range of many QCT and physiologic values overlapped considerably between the selected and excluded groups. The percent severe emphysema (<ⴚ 960 Hounsfield units [HU]), upper/ lower lung emphysema ratio (ⴚ 900 HU threshold), and residual volume were the key variables in the model predicting selection decisions (model r2 ⴝ 0.48; p < 0.0001). The model correctly predicted selection decisions in 87% of all cases, 91% of the selected group, and 78% of the excluded group. Surgical patients with a higher model-derived probability of selection had greater postoperative improvement in FEV1 and 6-min walk distance. Conclusions: Radiologic selection criteria are applied consistently to the majority of patients. QCT features are strongly associated with selection decisions, are related to outcome, and may help improve consistency and confidence in patient selection. (CHEST 2000; 117:991–998) Key words: CT; emphysema; lung volume reduction surgery Abbreviations: HU ⫽ Hounsfield units; LVRS ⫽ lung volume reduction surgery; NS ⫽ not significant; QCT ⫽ quantitative CT; RV ⫽ residual volume; TLC ⫽ total lung capacity; 6MW ⫽ 6-min walk distance

selection for lung volume reduction surP atient gery (LVRS) involves evaluation of multiple factors impacting inclusion and exclusion.1,2 Patients with debilitating symptoms not palliated by medical therapy, severe obstructive physiology, anatomic fea*From the Mallinckrodt Institute of Radiology (Drs. Gierada, Pilgram, Slone, and Mr. Villanueva), and the Divisions of Pulmonary and Critical Care Medicine (Drs. Lefrak and Yusen), and Cardiothoracic Surgery (Dr. Cooper), Washington University School of Medicine, St. Louis, MO 63110. Supported in part by the American Lung Association of Eastern Missouri.

tures appropriate for LVRS (hyperinflation and surgical “target areas” of severe emphysema), and lack of significant comorbid medical conditions are considered potential candidates. The generally favorable clinical outcome of patients following surgery3–5 Manuscript received July 26, 1999; revision accepted October 5, 1999. Correspondence to: David S. Gierada, MD, Mallinckrodt Institute of Radiology, Barnes-Jewish Hospital, 216 South Kingshighway Blvd, St. Louis, MO 63110; e-mail: gieradad@mir. wustl.edu

CHEST / 117 / 4 / APRIL, 2000

991

indicates the overall success of current selection strategies that rely on history and physical examination, physiologic testing, and imaging including chest radiography, CT, and perfusion scintigraphy. Nevertheless, many aspects of patient selection remain largely subjective and often difficult, particularly the interpretation of imaging studies for the severity and distribution of emphysema. The importance of anatomic features has been demonstrated in studies showing better clinical outcome after LVRS in patients with a heterogeneous distribution of emphysema that predominates in the upper lobes.6 –9 A large percentage of patients evaluated for LVRS are excluded because they lack this pattern,3,9 and instead have a uniform distribution of emphysema throughout the lungs, or emphysema that is only mild or moderate in severity on imaging studies. However, these anatomic features are quite variable and frequently difficult to categorize, and individual selection decisions are accompanied by varying degrees of uncertainty and confidence. In clinical practice, anatomic features are assessed subjectively by visual inspection of the imaging studies. Objective assessment is also possible, using whole-lung quantitative CT analysis (QCT). QCT measurements of lung density accurately reflect macroscopic emphysema.10,11 This technique has been used to assess the anatomic effects of LVRS12,13 and to identify preoperative anatomic features related to outcome after LVRS.14 Specific QCT guidelines for prospective use in selection have not yet been defined. The objectives of this study were to use QCT to see whether subjective assessments of radiologic features are applied consistently, and to derive objective selection criteria based on QCT measurements, physiologic data, and the selection decisions that resulted from our standard preoperative evaluation. To do this, we compiled the numerous QCT and physiologic variables considered in the preoperative assessment, and compared them in selected and excluded patients. Using multiple regression analysis to identify the variables having the greatest impact on selection decisions, we developed a model for patient selection based only on objective, quantifiable parameters. We then assessed the relevance of the model to postoperative outcome.

inappropriate severity and distribution of emphysema was listed as the major reason for exclusion. These decisions had been based on review of the chest radiographs, CTs, and perfusion scintigrams by a pulmonologist and/or thoracic surgeon, with occasional consultation by a thoracic radiologist. Fifty-two of these patients had chest CTs performed at our institution and available for QCT analysis. From a detailed review of clinic charts by two of the authors, we found exclusionary factors in 20 patients, such as prior pleurodesis, unresectable lung cancer, marked hypercapnea, and overall poor physical condition. Except for a pattern of emphysema on imaging studies that was judged unfavorable for LVRS, the remaining 32 patients were otherwise operable candidates, and formed the excluded group in this study. The selected group consisted of 70 patients selected for LVRS using criteria previously described,1 including review of radiologic studies as noted above, who had undergone LVRS between December 1993 and May 1995. These patients are from the first 142 patients to undergo LVRS at our institution, for whom CTs were available for QCT analysis. The total study group thus consisted of 102 patients: 51 women and 51 men, between the ages of 33 and 77 years (mean, 62 ⫾ 8 years). CT Scanning and Analysis The preoperative CTs were performed without IV contrast during full inspiration. Thirty were performed on a Somatom Plus 4 CT scanner (Siemens Medical Systems; Iselin, NJ) using spiral technique with 0.75-s scan time, 8-mm section thickness, and 8-mm/s table incrementation; 72 were performed on a Somatom Plus S scanner (Siemens Medical Systems) using incremental technique, 1-s scan time, and 8-mm (n ⫽ 10) or 10-mm (n ⫽ 62) section thickness. Whole-lung QCT analysis was performed at the scanner operator’s console using the Pulmo software option (Siemens Medical Systems). This provides semiautomated segmentation of the lung in each image and automated calculation and display of pixel density statistics. The QCT variables in this study were defined to be analogous to the anatomic features considered important in visual assessment,6,15 and are listed in Table 1. We used three indexes of global emphysema severity, three indexes of regional emphysema severity, two indexes of the heterogeneity in emphysema severity, and one index of the amount of reserve lung tissue (volume of lung minimally affected by emphysema). Physiologic Testing Spirometry and plethysmography were performed using a Medgraphics System 1085 (Medical Graphics; St. Paul, MN). Values reported were obtained following aerosolized albuterol administration. Arterial blood gas analysis was performed using a Model BG3 blood gas analyzer (Instrumentation Laboratory; Lexington, MA). Diffusing capacity for carbon monoxide was performed using the single-breath technique. Six-minute walk distance (6MW) was performed in a standardized manner, allowing as many stops and as much supplemental oxygen as needed. Interventions

Materials and Methods Patients To identify excluded patients appropriate for this study, a clinic database containing summary data of patients evaluated for but excluded from LVRS between July 1994 and March 1997 was searched. The search identified 103 patients for whom an 992

Patients selected for surgery had bilateral LVRS to reduce the volume of each lung by 20 to 30%, as previously described.15 Lung was resected in areas that showed the most severe emphysema on imaging studies and on direct inspection during surgery. Statistical Analysis QCT and physiologic test values in the selected and excluded patients were compared with two-tailed, unpaired t tests and dot Clinical Investigations

Table 1–QCT Parameters of Emphysema Relevant to the Preoperative Assessment for LVRS Anatomic Features Global emphysema severity

QCT Variables

Method of Determination

Mean lung density

Mean of all lung pixel density values

Percent emphysema

Percent of all lung pixels with density value ⬍ ⫺ 900 HU Percent of all lung pixels with density value ⬍ ⫺ 960 HU Percent of all pixels in the upper half of the lungs with density value below ⫺ 900 HU Percent of all pixels in the lower half of the lungs with density value below ⫺ 900 HU Ratio of upper to lower lung percent emphysema Standard deviation of all pixel density values in HU

Percent severe emphysema Regional emphysema severity

Upper-lung percent emphysema Lower-lung percent emphysema Upper/lower lung emphysema

Heterogeneity of emphysema severity

Standard deviation of the mean lung density Full width at half maximum

Reserve lung tissue

Volume of nonemphysematous lung tissue

plots, using Excel 4.0 software (Microsoft; Redmond, WA). Preoperative and postoperative physiologic test values were compared using two-tailed, paired t tests. Logistic regression was performed using JMP software (SAS Institute; Cary, NC). QCT and physiologic parameters were entered as individual independent variables, and the decision to operate as the dependent variable. Backwards stepwise multivariate logistic regression was performed using those parameters having the strongest association with the decision to operate. The resulting model was internally validated by testing it on the patients from whom it was derived. The effects of potentially significant variables that the model rejected were tested by forcing them into the model. The relevance of the model to postoperative clinical outcome was assessed in two ways. First, the outcome of surgical patients for whom the model predicted exclusion from LVRS was compared with the outcome of surgical patients for whom the model predicted selection. Second, outcome of surgical patients having the highest model-calculated probability of selection was compared with the outcome of the other surgical patients. Postoperative improvements in FEV1, Pao2, and 6MW, 6 months after surgery, were used as measures of outcome. If 6-month data were unavailable, 3-month data were used (eight patients). One-month data were used in one patient lacking 3- and 6-month data. Patients who died prior to postoperative pulmonary function testing were excluded from this outcome analysis.

Results Comparison of QCT and physiologic profiles of the excluded and selected patients is shown in Table 2. By QCT analysis, emphysema was more severe globally, and regionally in the upper lungs in the selected patients compared to the excluded patients. The greatest differences between the two groups were in the percent severe emphysema (⫺ 960 Hounsfield units [HU] threshold) and in the upper/ lower lung emphysema ratio (⫺ 900 HU threshold). Although statistically significant, the differences in

Width of the density histogram in HU at half the maximum height Volume of lung with density values between ⫺ 850 and ⫺ 701 HU

mean lung density, percent emphysema (⫺ 900 HU threshold), and standard deviation of the mean lung density were small. The difference in percent emphysema (⫺ 900 HU threshold) reflected a difference in the upper lungs; upper-lobe predominance of emphysema was greater in the selected group, while the percent emphysema in the lower lungs was the same in both groups. There was substantial overlap in characteristics of the two groups, even for the QCT variables that showed the greatest differences (Fig 1, top left and top right). No operable patient with ⬎ 30% severe emphysema or an upper/ lower lung emphysema ratio ⬎ 1.6 had been excluded from LVRS, and no patients with ⬍ 9% severe emphysema were selected. Comparison of physiologic test results revealed that all patients had severe obstructive lung disease, markedly increased lung volumes, impaired gas exchange, and low exercise capacity (Table 2). Abnormalities of all variables except for FEV1/FVC, residual volume (RV)/total lung capacity (TLC), Paco2, and 6MW were slightly more severe in the selected patients. Substantial overlap was seen, however, even for the variables having the greatest mean differences (Fig 1, bottom left and bottom right). The excluded group was older, but again, there was substantial overlap. No operable patient with a RV ⬎ 310% of predicted was excluded. Univariate logistic regression revealed statistically significant associations (p ⬍ 0.05) between the decision to operate and the variables for which mean differences between the two groups were statistically significant, but in most cases the association was weak. The five variables having ␹2 ⬎ 12 (p ⬍ 0.001) CHEST / 117 / 4 / APRIL, 2000

993

Table 2–QCT and Physiologic Profiles Compared in Patients Selected for and Excluded From LVRS* Variables

Selected Patients (n ⫽ 70)

Excluded Patients (n ⫽ 32)

p Value†

⫺ 895 ⫾ 13 95 ⫾ 7 93 ⫾ 32 62 ⫾ 8 27 ⫾ 9 72 ⫾ 11 52 ⫾ 14 1.5 ⫾ 0.5 1.1 ⫾ 0.3

⫺ 886 ⫾ 13 88 ⫾ 4 95 ⫾ 14 57 ⫾ 9 13 ⫾ 8 61 ⫾ 12 53 ⫾ 10 1.2 ⫾ 0.3 1.1 ⫾ 0.3

0.001 ⬍ 0.0001 NS 0.003 ⬍ 0.0001 ⬍ 0.0001 NS ⬍ 0.001 NS

QCT profile Mean lung density, HU SD of mean lung density, HU Full width at half maximum, HU % emphysema (⬍ ⫺ 900 HU) % severe emphysema (⬍ ⫺ 960 HU) % emphysema, upper lung (⬍ ⫺ 900 HU) % emphysema, lower lung (⬍ ⫺ 900 HU) Upper/lower-lung emphysema Normal density lung (⫺ 850 to ⫺ 701 HU), L Physiologic profile Age, yr FEV1, % predicted FVC, % predicted FEV1/FVC TLC, % predicted RV, % predicted RV/TLC Pao2, mm Hg Paco2, mm Hg Dlco, % predicted 6 MW, feet

61 ⫾ 8 23 ⫾ 6 69 ⫾ 17 0.28 ⫾ 0.07 144 ⫾ 19 288 ⫾ 59 0.71 ⫾ 0.07 59 ⫾ 10 44 ⫾ 8 32 ⫾ 13 1,110 ⫾ 270

65 ⫾ 7 27 ⫾ 7 81 ⫾ 29 0.28 ⫾ 0.06 136 ⫾ 17 242 ⫾ 35 0.69 ⫾ 0.07 64 ⫾ 9 44 ⫾ 7 36 ⫾ 15 990 ⫾ 270

0.01 0.002 0.01 NS 0.05 0.0001 NS 0.02 NS NS 0.03

*Values are presented as mean ⫾ SD; Dlco ⫽ diffusing capacity of the lung for carbon monoxide. †p values ⬎ 0.05 are not statistically significant.

and r2 ⬎ 0.10 were used in multivariate logistic backwards stepwise regression. These included the percent severe emphysema (⫺ 960 HU threshold; r2 ⫽ 0.35), the upper-lung percent emphysema (⫺ 900 HU threshold; r2 ⫽ 0.15), the upper/lowerlung emphysema ratio (⫺ 900 HU threshold; r2 ⫽ 0.11), the standard deviation of the mean lung density ( r2 ⫽ 0.17), and the percent predicted RV ( r2 ⫽ 0.13). This produced a model in which the combination of percent severe emphysema (⫺ 960 HU threshold), upper/lower-lung emphysema (⫺ 900 HU threshold), and percent predicted RV explained nearly half of the variability in the selection decisions ( r2 ⫽ 0.48; Table 3). The standard deviation of the mean lung density and the upperlung percent emphysema did not make a significant contribution in the multivariate model. For internal validation, the model was tested on the patients used to create it. By entering individual patient values for each of the three variables, the model generates a probability of selection in the form of a proportion. We considered probabilities ⬍ 0.5 as predictive of exclusion, and probabilities ⱖ 0.5 as predictive of selection. Predictions corresponded to actual decisions in 87% of all patients. The model predicted acceptance for 7 of the 32 patients who were excluded (22%), and predicted exclusion for 6 of the 70 patients who underwent LVRS (9%). 994

Physiologic changes in the patients who underwent LVRS are listed in Table 4. Five patients died, for a hospital mortality rate of 7%, and these patients were excluded from the outcome comparison. Comparison of physiologic outcome after LVRS in operated patients for whom the model predicted exclusion vs outcome of those for whom the model predicted selection revealed a trend toward greater physiologic improvement in the patients for whom the model predicted selection (Table 5). Significantly greater mean improvement in FEV1 and 6MW was found for patients having model-derived selection probabilities higher than thresholds from 0.85 to 0.98 (outcome for patients with a probability of selection ⬍ 0.9 and ⬎ 0.9 is shown in Table 5). Correlations between the probability of selection and postoperative improvement in FEV1 (r ⫽ 0.34; p ⫽ 0.0048) and 6MW (r ⫽ 0.25; p ⫽ 0.048) were not high. However, when patients were stratified into quartiles according to the model-derived probability of selection, greater improvement in FEV1 and 6MW was seen in patients with a higher probability (Table 6). To test the impact of other potentially significant clinical variables, we forced age, gender, percent predicted FEV1, and 6MW into the model. Age was not significant (NS; ␹2 ⫽ 0; p ⫽ 0.98), and when removed, the three other added variables became significant (␹2 ⫽ 4 to 6; p ⫽ 0.01 to 0.04), and the Clinical Investigations

Table 4 –Physiologic Data From 65 Patients Before and After LVRS* Variables

Before LVRS

After LVRS

% Change†

TLC, L‡ (% predicted) RV, L‡ (% predicted) FEV1, L (% predicted) FVC, L (% predicted) Pao2, mm Hg Paco2, mm Hg 6MW, feet

8.15 ⫾ 1.48 (143 ⫾ 19) 5.80 ⫾ 1.29 (286 ⫾ 62) 0.64 ⫾ 0.20 (23 ⫾ 6) 2.37 ⫾ 0.71 (69 ⫾ 17) 59 ⫾ 10 44 ⫾ 7 1,120 ⫾ 270

7.11 ⫾ 1.51 (124 ⫾ 19) 4.18 ⫾ 1.24 (205 ⫾ 57) 0.99 ⫾ 0.38 (36 ⫾ 12) 2.85 ⫾ 0.81 (84 ⫾ 18) 68 ⫾ 11 40 ⫾ 6 1,280 ⫾ 350

⫺ 13 ⫺ 28 55 20 15 ⫺9 14

*Values are presented as mean ⫾ SD; excludes five patients who died prior to postoperative testing. †All changes are p ⬍ 0.0001, two-tailed paired t test. ‡TLC and RV not available after surgery in three patients.

accepted, and acceptance predicted for eight of the patients who were excluded. Discussion

Figure 1. Dot plots demonstrate overlap in quantitative CT (top left, top right) and physiologic (bottom left, bottom right) test values in patients selected for and excluded from LVRS.

overall model containing six variables improved (␹2 ⫽ 75; p ⬍ 0.0001; r2 ⫽ 0.59). Internal validation revealed minimal improvement in overall accuracy (88%) compared to the three-variable model, with exclusion predicted for four of the patients who were

Table 3–Multiple Logistic Regression Model for Patient Selection* 95% Confidence Interval p Value

Variables

Odds Ratio

% severe emphysema (at 1% increments) Upper/lower emphysema (at 0.1 increments) RV % predicted (at 10% increments)

1.21

1.11–1.34

1.32

1.11–1.62

0.004

1.21

1.05–1.43

0.015

⬍ 0.0001

*Includes patients accepted for surgery (n ⫽ 70) and excluded from surgery (n ⫽ 32); model ␹2 ⫽ 61 (p ⬍ 0.0001); model r 2 ⫽ 0.48.

Despite increasing experience with LVRS, selection of operable patients (ie, need for palliation, severe obstructive lung disease primarily due to emphysema rather than intrinsic airway disease, elevated lung volumes, and no exclusionary comorbid medical conditions) remains one of the greatest challenges. Attempts to improve patient selection have primarily focused on identifying the preoperative anatomic and physiologic parameters that correlate with various measures of postoperative outcome.7,9,14,16,17 Recognizing that most patients selected have had a favorable outcome, we directed our attention in this study toward a quantitative analysis of the otherwise operable candidates who were excluded because of inappropriate anatomic features of emphysema noted on subjective assessment. Use of QCT allowed assessment of whether patient selection based on the subjective visual assessment of imaging studies has been consistent. We also were able to determine which CT and physiologic variables had the greatest impact on selection decisions, and whether these variables are related to postoperative outcome. This allowed modeling of the clinical decision-making process using objective, quantitative data, which may serve as a step toward the development of practical, objective selection guidelines. The QCT comparison of selected and excluded patients confirms that overall, the subjective visual assessment of radiologic studies resulted in selection of the patients with more severe emphysema, and CHEST / 117 / 4 / APRIL, 2000

995

Table 5–Relationship Between Quantitative Selection Model and Postoperative Outcome* Mean Postoperative Change in Variables

% Severe Emphysema

Upper/Lower Emphysema

% Predicted RV

FEV1, L

Pao2, mm Hg

6MW, feet

15 ⫾ 4

1.1 ⫾ 0.3

260 ⫾ 40

0.21 ⫾ 0.17

5⫾7

40 ⫾ 161

27 ⫾ 9

1.6 ⫾ 0.5

290 ⫾ 60

0.35 ⫾ 0.28

9 ⫾ 11

170 ⫾ 270

0.002

0.02

0.17

0.22

0.42

0.24

20 ⫾ 6

1.2 ⫾ 0.4

260 ⫾ 50

0.19 ⫾ 0.17

8⫾7

60 ⫾ 220

30 ⫾ 9

1.7 ⫾ 0.5

310 ⫾ 60

0.43 ⫾ 0.28

9 ⫾ 13

220 ⫾ 270

⬍ 0.0001

0.0002

0.003

0.0002

0.54

0.02

By model prediction† Exclusion predicted (n ⫽ 6) Selection predicted (n ⫽ 59) p value‡ By probability of selection Probability ⬍ 0.9 (n ⫽ 20) Probability ⱖ 0.9 (n ⫽ 45) p value‡

*Values are presented as mean ⫾ SD and exclude five patients who died prior to postoperative data collection. †Model output is in the form of a proportion corresponding to the probability of selection; exclusion was predicted for those with probability ⬍ 0.5, and selection predicted for those with probability ⱖ 0.5. ‡Two-tailed t test comparing mean values.

greater upper-lobe predominance of emphysema. The differences between the two groups in functional impairment shown by physiologic testing likely relate to these differences in emphysema severity found by QCT. Although the excluded group also had substantial emphysema by QCT, it was not quite as severe as in the selected patients, and the upperlobe predominance was not as pronounced. However, the moderate amount of overlap between the selected and excluded groups suggests that radiologic criteria may not be applied consistently in some cases. This is understandable, considering that visual assessment is subjective. By multiple regression modeling, the individual selection decisions in operable patients were accurately predicted from objective, quantitative data. An encouraging aspect of the model is that each component is analogous to a feature considered important in selection, according to criteria derived in part

Table 6 –Postoperative Physiologic Outcome in Selection Model-Derived Probability Quartiles* Mean Postoperative Change in

Quartile

n

Probability of Selection*

FEV1, L

6MW, feet

1 2 3 4

17 16 16 16

0–0.787 0.817–0.941 0.943–0.994 0.994–0.999

0.20 0.28 0.38† 0.50‡

70 100 180 290†

*Excluding five patients who died prior to postoperative data collection. †p ⬍ 0.05 compared with quartile 1; two-tailed, unpaired t test. ‡p ⬍ 0.01 compared with quartile 1 and p ⬍ 0.05 compared with quartile 2; two-tailed, unpaired t test. 996

from studies assessing the relationship between preoperative variables and outcome.7,9 This suggests that QCT data could be a valuable adjunct for clinical use, potentially improving consistency and confidence in patient selection. In the simplest model containing three variables (r2 ⫽ 0.48), the combination of the percent severe emphysema (⫺960 HU threshold) and the upper/ lower lung emphysema distribution (⫺900 HU threshold) accounted for most of the variability in selection decisions (r2 ⫽ 0.42). Of these, the percent severe emphysema had the greatest influence (r2 ⫽ 0.35). RV was also a significant independent factor in the model, suggesting that the degree of hyperinflation was important in selection. The smaller influence of this variable may be due to a lack of variability among the candidates for LVRS, reflecting our approach of restricting evaluations to patients that have at least modest hyperinflation. The more complex six-variable model accounted for a higher proportion of the variability in selection decisions (r2 ⫽ 0.59), but resulted in only marginal improvement in the number of selection decisions correctly predicted. The relationship between the model and postoperative physiologic outcome helps to validate both the subjective selection criteria used in LVRS and the quantitative model derived here. In stratified groups, patients with the highest probability of selection calculated by the model had substantially and significantly greater postoperative physiologic improvement. There was also a trend toward lesser physiologic improvement in the small number of patients operated on for whom the model predicted exclusion. The lack of a high correlation between the Clinical Investigations

probability of selection and postoperative physiologic outcome is likely due to the small range of probability values seen among the selected patients, as nearly two-thirds of those selected had a selection probability of ⱖ 0.90. Of the two emphysema threshold density values evaluated, the ⫺ 960 HU threshold had greater relevance to patient selection. The amount of severe emphysema defined by this threshold was the parameter that best discriminated between the excluded and selected groups in the comparison of mean values, and had the greatest impact in the multivariate model. This parameter also had the strongest correlation with clinical outcome after LVRS among multiple QCT and physiologic parameters in a previous study.14 Although the difference between selected and excluded groups in percent emphysema based on the ⫺ 900 HU threshold was also statistically significant (p ⬍ 0.01), the difference was not very large. Thus, while thresholds closer to ⫺ 900 HU on scans obtained with conventional section thickness have been found to correlate well with macroscopic emphysema in pathologic specimens,18 the ⫺ 960 HU threshold appears more appropriate for evaluating global emphysema severity using conventional CT section thickness in LVRS candidates. Of the two QCT variables used to indicate the variation in lung density values, only the standard deviation of the mean density had a significant association with selection decisions by univariate logistic regression; this variable and the percent upper-lung emphysema dropped out of the final model, suggesting that they are influenced by one or more of the other stronger variables. Neither the standard deviation of the mean density nor the full width at half maximum of the density histogram reflect the spatial distribution of emphysema, however, and therefore they cannot be considered directly analogous to the degree of emphysema heterogeneity, ie, the unevenness in the spatial distribution of emphysema, or presence of surgical target areas. Although likely related in part to the upper/lower lung emphysema distribution, a quantitative measure of spatial heterogeneity, such as through texture analysis,19 or quantitation of continuous low-attenuation areas,20 might improve the model. A shortcoming of the patient selection model developed here is the lack of a “gold standard”; it is unknown whether the clinical decisions to exclude patients were appropriate. The model predicted selection for 7 of the 32 excluded patients, indicating that 22% of excluded patients had relevant features similar to those of selected patients, suggesting that at least some excluded patients might benefit from LVRS. This could only be evaluated by operating on

these patients. Of additional interest is that the model predicted exclusion for six of the selected patients. Since most of these patients obtained at least mild physiologic benefit, some excluded patients for whom exclusion was predicted by the model also might conceivably obtain some physiologic benefit from LVRS. This is supported by studies showing a lesser degree of but definite clinical improvement after LVRS in patients with a homogeneous distribution of emphysema throughout the lungs.8 However, the likelihood of physiologic improvement appears greater for those with higher model-derived probabilities of selection. This study did not include some of the variables that were used prospectively in patient selection decisions. Radiographic hyperinflation, perfusion scintigraphy, self-reported dyspnea scores, and even the physical appearance of the patient are all factors that were undoubtedly considered to some degree, but were not a part of the quantitative model. Despite this, the modeling process was relatively successful. Including those additional features that could be quantified objectively might improve the model. The retrospective identification of patients for the excluded group, which relied on review of written records, may have introduced bias into the study. However, the excluded patients were identified prior to the quantitative analysis of their CT studies. In summary, QCT analysis of lung density patterns shows that subjective radiologic evaluation resulted in selection of patients for LVRS in whom emphysema was more severe and had a greater upper-lung distribution compared to those excluded. The percent severe emphysema (⫺ 960 threshold), upperlower-lung emphysema ratio (⫺ 900 HU threshold), and RV (percent of predicted) were the variables having the greatest impact on selection decisions. An objective selection model based on these variables accurately predicted the outcome of the selection process, and relates to postoperative physiologic outcome. Overlap of characteristics among selected and excluded patients and model prediction of selection of some excluded patients suggests that additional potentially appropriate patients may be identified using QCT. These results help support the validity of QCT as a potential tool for patient selection in LVRS, and suggest that QCT could help to improve consistency and confidence in this often difficult aspect of LVRS. Prospective application of this model to subsequent LVRS candidates would be helpful in further testing its validity. Identification of other preoperative variables that are associated with postoperative outcome in addition to those currently considered in the preoperative assessment, such as inspiratory presCHEST / 117 / 4 / APRIL, 2000

997

sure,21 may refocus patient selection strategies over time. Such potential improvements to patient selection could also be incorporated into the type of model presented here. ACKNOWLEDGMENT: The authors thank Mary Pohl, RN, for patient database support; Jamie Reynolds, Mary Vogel, and Claudia Phillips for assistance with retrieval of clinical records and patient scheduling; and the CT technology staff of the Mallinckrodt Institute of Radiology for retrieval of electronically archived CT examinations. We acknowledge the vital efforts of all other contributors to the success of the LVRS program, including Gail Davis, RN, nurse coordinator; Dottie Biggar, RN, and the pulmonary rehabilitation staff; the nursing service of the Division of Cardiothoracic Surgery; the chest physiotherapists and respiratory therapists; the cardiothoracic anesthesia staff; and the house officers and other clinical staff who are all essential to the complicated care required of these patients.

References 1 Yusen RD, Lefrak SS. Evaluation of patients with emphysema for lung volume reduction surgery. Washington University Emphysema Surgery Group. Semin Thorac Cardiovasc Surg 1996; 8:83–93 2 Brenner M, Yusen R, McKenna R, Jr, et al. Lung volume reduction surgery for emphysema. Chest 1996; 110:205–218 3 Cooper JD, Patterson GA, Sundaresan RS, et al. Results of 150 consecutive bilateral lung volume reduction procedures in patients with severe emphysema. J Thorac Cardiovasc Surg 1996; 112:1319 –1329 4 Kotloff RM, Tino G, Bavaria JE, et al. Bilateral lung volume reduction surgery for advanced emphysema: a comparison of median sternotomy and thoracoscopic approaches. Chest 1996; 110:1399 –1406 5 Brenner M, McKenna RJ, Gelb AF, et al. Dyspnea response following bilateral thoracoscopic staple lung volume reduction surgery. Chest 1997; 112:916 –923 6 Slone RM, Gierada DS. Radiology of pulmonary emphysema and lung volume reduction surgery. Semin Thorac Cardiovasc Surg 1996; 8:61– 82 7 Slone RM, Pilgram TK, Gierada DS, et al. Lung volume reduction surgery: comparison of preoperative radiologic features and clinical outcome. Radiology 1997; 204:685– 693 8 Weder W, Thurnheer R, Stammberger U, et al. Radiologic emphysema morphology is associated with outcome after surgi-

998

cal lung volume reduction. Ann Thorac Surg 1997; 64:313–319 9 McKenna RJ, Jr, Brenner M, Fischel RJ, et al. Patient selection criteria for lung volume reduction surgery. J Thorac Cardiovasc Surg 1997; 114:957–964 10 Thurlbeck WM, Muller NL. Emphysema: definition, imaging, and quantification. AJR Am J Roentgenol 1994; 163:1017–1025 11 Stern EJ, Frank MS. CT of the lung in patients with pulmonary emphysema: diagnosis, quantification, and correlation with pathologic and physiologic findings. AJR Am J Roentgenol 1994; 162:791–798 12 Holbert JM, Brown ML, Sciurba FC, et al. Changes in lung volume and volume of emphysema after unilateral lung reduction surgery: analysis with CT lung densitometry. Radiology 1996; 201:793–797 13 Bae KT, Slone RM, Gierada DS, et al. Patients with emphysema: quantitative CT analysis before and after lung volume reduction surgery; work in progress. Radiology 1997; 203: 705–714 14 Gierada DS, Slone RM, Bae KT, et al. Pulmonary emphysema: comparison of preoperative quantitative CT and physiologic index values with clinical outcome after lung-volume reduction surgery. Radiology 1997; 205:235–242 15 Cooper JD, Patterson GA. Lung-volume reduction surgery for severe emphysema. Chest Surg Clin North Am 1995; 5:815– 831 16 Wang SC, Fischer KC, Slone RM, et al. Perfusion scintigraphy in the evaluation for lung volume reduction surgery: correlation with clinical outcome. Radiology 1997; 205:243– 248 17 Brenner M, McKenna R Jr, Gelb A, et al. Objective predictors of response for staple versus laser emphysematous lung reduction. Am J Respir Crit Care Med 1997; 155:1295–1301 18 Muller NL, Staples CA, Miller RR, et al. “Density mask”: an objective method to quantitate emphysema using computed tomography Chest 1988; 94:782–787 19 Uppaluri R, Mitsa T, Sonka M, et al. Quantification of pulmonary emphysema from lung computed tomography images. Am J Respir Crit Care Med 1997; 156:248 –254 20 Sakai N, Mishima M, Nishimura K, et al. An automated method to assess the distribution of low attenuation areas on chest CT scans in chronic pulmonary emphysema patients. Chest 1994; 106:1319 –1325 21 Ingenito EP, Evans RB, Loring SH, et al. Relation between preoperative inspiratory lung resistance and the outcome of lung-volume-reduction surgery for emphysema. N Engl J Med 1998; 338:1181–1185

Clinical Investigations