Immediate Results of Percutaneous Mitral Commissurotomy
A Predictive Model on a Series of 1514 Patients
Background The wide use of percutaneous mitral commissurotomy (PMC) underlines the need to identify the predictive factors of the results. Using a large series allowed us to develop a multivariate model that can be applied to improve patient selection.
Methods and Results Between 1986 and 1995, PMC was undertaken in 1514 patients. Mean age was 45±15 years. Echocardiography showed that 245 patients (16%) had pliable valves and mild chordal thickening (group 1), 886 (59%) had extensive subvalvular disease (group 2), and 383 (25%) had calcified valves (group 3). PMC failed in 22 patients; it was performed with a single balloon in 30 patients, a double balloon in 586, and the Inoue balloon in 876. Good immediate results were defined as a valve area ≥1.5 cm2 with mitral regurgitation Sellers' grade ≤2 and were obtained in 1348 patients (89%). A logistic model developed from the first 1088 cases identified the following predictors of immediate results: age (P=.004), echocardiographic group (P<.0001), valve area (P<.0001), and effective balloon dilating area (EBDA) (P=.03). Two interactions were significant: age at previous commissurotomy (P=.013) and EBDA by initial mitral regurgitation (P=.034). The type of balloon was of borderline significance (P=.09). The model was validated on an independent sample comprising the subsequent 426 procedures. For a threshold of probability of good results of .75, sensitivity was 92%, specificity 25%, and predictive accuracy 87%.
Conclusions Prediction of the immediate results of PMC is multifactorial. The predictive model developed and validated can be contributive in decision making for individual patients.
Since 1984,1 numerous reports have demonstrated the safety and the efficacy of percutaneous mitral commissurotomy (PMC),2 3 4 5 6 and this technique has replaced closed commissurotomy for the treatment of mitral stenosis in young patients with pliable valves. In western countries, however, PMC is often used in older patients with more severe valve disease who are theoretically less good candidates for balloon commissurotomy. This wide use of PMC stresses the need to identify the predictive factors of the immediate results to improve the selection of candidates for this technique. Analyses of the predictors of immediate results of PMC have been performed in different single- or multicenter series.3 4 5 6 In multicenter series, however, differences between centers, in particular in patient selection, grading of baseline characteristics, and the experience of the operators, may influence the identification of the predictive factors.
The present study reports the immediate results in a large single-center series with a variety of patient subsets undergoing PMC with the most frequently used techniques. From this series, a predictive model, including interactions, was developed and then validated prospectively on an independent sample. This validation allowed the evaluation of the fit of the model and its applicability in decision making by discriminant analysis.
From March 1986 to March 1995, PMC was indicated in 1514 consecutive patients who had mitral stenosis with a valve area <1.5 cm2. Their mean age was 45±15 years (range, 9 to 86 years); 222 patients (15%) had had previous commissurotomy a mean of 15±8 years before PMC (closed commissurotomy in 165, open commissurotomy in 40, PMC in 17). Contraindications were calcification of both commissures; left atrial thrombus on transesophageal echocardiography, which was systematically performed except in the first 180 procedures; mitral regurgitation Sellers' grade >2; coexistence of aortic valve disease with a valve area ≤0.8 cm2 or aortic regurgitation Sellers' grade ≥3; or coronary stenosis ≥70% of diameter. A predictive model was initially developed from the first 1088 procedures and then validated prospectively on the sample of the 426 subsequent cases.
Dilatation was performed in all cases by the antegrade transvenous approach. The mitral valve was dilated with a single balloon in 30 patients (Trefoil, 3×12 mm) and with a double balloon in 586 (Trefoil, 3×10 or 3×12 mm plus 15 or 19 mm according to the patient's height).2 After October 1990, the Inoue balloon was systematically used in 876 patients, with stepwise inflation under echocardiographic guidance. Balloon size was chosen according to the patient's height, and the balloon was then inflated in steps of 1 or 2 mm. Valve area, commissural splitting, and the degree of mitral regurgitation were assessed by transthoracic echocardiography after each inflation. Our criteria for stopping the procedure were complete opening of at least one commissure with a valve area >1 cm2/m2 body surface area or >1.5 cm2 or the appearance or increase of regurgitation >1/4.7
After echocardiography and fluoroscopy, valve anatomy was classified into three groups (Table 1)⇓ as previously described.2 7 8 The pliability of the anterior leaflet was assessed by two-dimensional echocardiography in the parasternal long-axis view. The length of chordae was measured in apical four- and two-chamber views. The presence of any calcification on fluoroscopy was necessary to classify patients into group 3 because of the limitations of echocardiography in differentiating nodular fibrosis from calcification. To indicate the correspondence with the Wilkins score, both scores were evaluated in a subset of 40 patients and showed that the mean±SD Wilkins score was 8.0±0.8 (range, 7 to 9) for echocardiographic group 1, 9.9±1.3 (range, 8 to 12) for group 2, and 12.5±1.3 (range, 10 to 15) for group 3.
Echocardiographic examination was performed in the same laboratory by three experienced operators on the day preceding PMC and 24 to 48 hours after the procedure. The reference measurement for valve area for statistical analysis was planimetry by two-dimensional echocardiography. Planimetry was not feasible in 54 patients (4%) before PMC and in 97 (6%) after PMC. In these cases, the Doppler pressure half-time was used instead. For the whole series, the correlation between the assessment of valve area by planimetry and by the pressure half-time was r=.77 before the procedure and r=.81 after the procedure. Before and after the procedure, right and left heart catheterization was performed, cardiac output was determined by the thermodilution method, and the degree of mitral regurgitation was assessed according to Sellers' classification on left ventriculography in a 30° right anterior oblique view. The degree of regurgitation could not be assessed on ventriculography in 97 patients (6%), so the analysis used an assessment by color Doppler; its concordance with ventriculography in the whole series was κ=.65. Mitral regurgitation, which was assessed only by echo-Doppler, was always <2/4. The effective balloon dilating area (EBDA) was calculated from the usual formula for the double-balloon technique9 and from the diameter at the last inflation for the Inoue balloon. All data were entered prospectively into a computerized database.
Quantitative variables were expressed as mean±SD. When included in a logistic model, they were divided into subgroups with clinically chosen cutoff points to simplify the interpretation of odds ratios. The dependent variable was a composite end point of good immediate results that associated a mitral valve area ≥1.5 cm2 and a regurgitation ≤2/4. Analysis of the predictors of immediate results was based on the first 1088 procedures of PMC (March 1986 to February 1993). The effects associated with the learning curve were studied by comparison of the results in each group of 50 consecutive patients with the results of all subsequent patients. Univariate analysis used the unpaired Student's t test for quantitative variables and the χ2 test with Yates' correction if necessary for qualitative variables. Univariate analysis included the 14 preprocedure variables listed in Tables 2⇓ and 3⇓ and two procedure-related variables, namely, the type of balloon and the EBDA. Variables with P<.25 were entered into a logistic model without interactions and were selected by a stepwise procedure using a significance level of P=.25 to enter or to remove variables.10 All two-way interaction terms between these selected variables and the two procedure-related variables were systematically studied and added in a second logistic model. Final selection of independent variables was made by a backward procedure with a significance level of P=.05 except for the “balloon” variable and the variables included in interaction terms, which were systematically maintained. Adjusted odds ratios were derived from the coefficients of the final multivariate logistic model.
This final logistic model can be used to predict the probability of good immediate results for any given patient according to the particular characteristics and procedure-related variables and was so used prospectively on the independent sample provided by the last 426 procedures (March 1993 to March 1995). The fit of the model was assessed by comparison of the predicted and observed numbers of good and inadequate immediate results for different classes of predicted probability of good results according to the method described by Lemeshow and Hosmer.11 The quality of the discrimination achieved with the logistic model was established for different thresholds of probability of good results by assessment of sensitivity (the percentage of good predicted results among good observed results), specificity (the percentage of inadequate predicted results among inadequate observed results), and predictive accuracy (the percentage of patients correctly classified by the model). The quality of the discrimination was measured by the area under the receiver-operating characteristic (ROC) curve according to the method using the Wilcoxon statistic.12
All analysis was performed with SAS statistical software (SAS Institute Inc, release 6.07), and estimation of the coefficients of the logistic model used the logistic procedure.
Description of Immediate Results
The clinical characteristics of the 1514 patients are detailed in Table 2. Of the 383 patients who had valve calcification (echocardiographic group 3), the extent of calcification was mild in 224, moderate in 145, and massive in 14. Technical failure occurred in 22 patients and was related to hemopericardium in 2 cases, embolism in 1, inability to cross the septum in 5, and inability to position the balloon correctly across the valve in 14. Dilatation was therefore effective in 1492 patients; hemodynamic and echocardiographic data before and after these procedures are shown in Table 3. Adverse events were in-hospital death in 6 patients (0.4%), hemopericardium with tamponade in 4 (0.3%), embolism leaving sequelae in 4 (cerebral 3, coronary 1) (0.3%), severe mitral regurgitation Sellers' grade ≥3 in 51 (3.4%), and local complications leading to vascular surgery in 12 (0.8%).
Good immediate results as defined by the composite end point were obtained in 1348 patients (89%). The 144 inadequate immediate results were related to insufficient valve opening (valve area <1.5 cm2) in 93 cases and severe mitral regurgitation (Sellers' grade ≥3) in 51. All cases of death, hemopericardium, and embolism with sequelae occurred in patients who had had inadequate immediate results of PMC.
Predictive Factors of Immediate Results
In the first 1088 procedures, technical failures were more frequent in the first 50 cases (16.0% versus 1.4% for later procedures, P<.0001). Eight of the 22 failed procedures of the whole series occurred among the first 50 patients. The frequency of inadequate results was also higher for the first 50 procedures than for later procedures (21.4% versus 10.1%, P=.02), but there was no significant difference between the 51st to 100th procedures and subsequent procedures (10% versus 12%, P=.78). Because of this learning effect, the first 50 procedures were excluded from the analysis of predictors. No learning effect was found for the procedures performed with the Inoue balloon: there were no failed procedures and no differences in inadequate results (10% for the first 50 versus 7.4% for subsequent procedures, P=.49). All procedures performed with the Inoue balloon were therefore included. Because the first 50 procedures and the 14 other failed procedures were not taken into account, analysis of predictors of the immediate results of PMC was finally carried out on the training sample of 1024 procedures, among which there were 103 inadequate results (10%).
By univariate analysis, only two variables were not linked to immediate results: sex (P=.55) and left atrial pressure (P=.33). Of the 12 patient-related variables shown to be significant in univariate analysis (Table 4⇓), the first logistic model showed 7 to be significant at P=.25: age, previous commissurotomy, echocardiographic group, mitral valve area, degree of mitral regurgitation, mean gradient, and functional class.
The second and final logistic model included 5 patient-related variables, 2 procedure-related variables, and 2 interactions. The strength of the predictive factors is indicated by the adjusted odds ratios in Table 5⇓. The risk of inadequate immediate results was increased in particular in patients with higher echocardiographic group (P<.0001), lower initial mitral valve area (P<.0001), and greater age (P=.004). The interaction between age and previous commissurotomy (P=.013) showed that the increase in the risk of inadequate results with greater age was more marked in patients who had had previous commissurotomy. Previous commissurotomy was not a predictor of inadequate results in itself (P=.65) but was a predictor only in patients >50 years old, as shown by the interaction with age. The other significant interaction concerned EBDA and mitral regurgitation (P=.034) and showed that immediate results were better with a large balloon only if there was no prior mitral regurgitation (EBDA, P=.03; mitral regurgitation, P=.65). There was a nonsignificant trend toward better immediate results with the Inoue balloon than with the double balloon (P=.09). No other interaction reached statistical significance, in particular that between the type of balloon and the other variables, as detailed in Table 6⇓.
Validation and Testing of the Predictive Model
This final logistic model can be used to calculate the predicted probability of good immediate results of PMC for any given patient (see “Appendix”). The model was so validated prospectively on the independent test sample comprising the 426 procedures performed between March 1993 and March 1995. The characteristics of these 426 patients representing the variables identified as predictors are shown in Table 7⇓. There were no failed procedures and 32 cases of inadequate immediate results in this sample (7.5%). The good fit of the model is shown by the absence of significant difference between predicted and observed immediate results in these 426 patients (χ2=9.46, df=6, P=.15).
The results of discriminant analysis performed with the final logistic model are shown in Table 8⇓. The high sensitivity for different thresholds of discrimination is indicative of good prediction of good immediate results. Conversely, the low specificity indicates that prediction of inadequate immediate results is weak. The combined variation of sensitivity and specificity is represented on the ROC curve (Figure⇓). The area under the ROC curve is 0.72 in the validation sample, and the application of the model to the training sample gives an area under the ROC curve of 0.75.
This single-center series comprising patients with varied characteristics demonstrates the efficacy of PMC with 90% of good immediate results. The multivariate model shows that the prediction of immediate results for PMC is multifactorial, based on anatomic, clinical, and procedural variables. The identification of interactions allows a better understanding of the relationships between the predictive variables and the results of PMC. The testing of the model on an independent sample validates the predictors identified. Moreover, this validation enables these predictors to be applied in practice as an aid to decisions for individual patients. The model allows the identification of ideal candidates for PMC but also demonstrates intrinsic limitations in the prediction of inadequate immediate results.
The composite end point of good immediate results used in our study took into account the two main phenomena that may require mitral surgery: insufficient valve opening and the occurrence of severe mitral regurgitation. A mitral valve area >1.5 cm2 generally provides normal hemodynamic conditions and leads to a persistent functional improvement, so this threshold has been used in most studies,4 5 6 13 14 even if clinical improvement may be observed in certain patients with a final valve area between 1.0 and 1.5 cm2. In the present study, mitral valve area was assessed by planimetry 24 to 48 hours after the procedure, because calculation by the Gorlin formula after PMC may be misleading, particularly in cases of interatrial shunt.15 Because Doppler estimation takes into account factors other than valve area, it may also give inaccurate results immediately after the procedure.16 Only planimetry allows direct measurement of valve area and may therefore be considered to be the reference measurement when performed by experienced operators.17 Nevertheless, planimetry cannot always be performed in patients with low echogenicity or with a very irregular, calcified mitral orifice. In such cases, we used Doppler measurements taken 1 or 2 days after PMC so as to avoid a selection bias caused by the elimination of patients in whom planimetry was not feasible, such as elderly persons or those with heavily calcified valves, who are often at high risk for inadequate immediate results. The occurrence of severe mitral regurgitation after PMC often compromises the short-term functional outcome of PMC.18 This justifies the inclusion of mitral regurgitation in the end point of immediate results.2 13 14 Our composite end point of immediate results does not take into account deaths, tamponades, or embolism-leaving sequelae, which were very infrequent and were always associated with inadequate immediate results of PMC in our series. Atrial shunts after antegrade PMC are generally small and usually decrease or disappear on follow-up19 and therefore are not included in our end point.
Predictive Factors of Immediate Results
In our series, as in others, the learning effect had an incidence on technical failure20 21 and the quality of immediate results,5 20 and for this reason, the first 50 procedures of our series were not included in the analysis of predictive factors to avoid the influence of this learning effect on the identification of the predictors.
In the present study, mitral anatomy is one of the most significant predictors of immediate results. Despite differences in the methods of echocardiographic assessment of mitral anatomy, many studies have identified anatomy as a predictive factor of mitral valve area2 3 4 5 6 14 or mitral regurgitation3 after PMC. In fact, mitral anatomy was initially considered to be the main predictor of the results of PMC, but it later appeared to be only a relative predictor. This latter view is partly explainable by limitations in echocardiographic evaluation of mitral anatomy, particularly the frequent underestimation of subvalvular disease and the absence of assessment of commissural areas. Another highly significant predictive factor of immediate results is mitral valve area before PMC, as has been reported in other studies.5 6 13
A distinctive feature of our study is the identification of predictive factors that are linked to the results through interactions. The increase in the risk of inadequate immediate results with age is greater in the case of previous commissurotomy, whatever its type. Previous commissurotomy is a predictor of inadequate immediate results only in patients >50 years old. The predictive value of age or previous commissurotomy has been demonstrated by univariate or multivariate analyses in other series3 5 6 22 but without reference to possible interactive effects. The predictive value of these factors in a multivariate model may suggest that the duration of rheumatic disease and previous commissurotomy may have damaged valve anatomy, even though this was not apparent on echocardiography. The second interaction concerns EBDA and the degree of mitral regurgitation before PMC. A higher valve area after PMC with a large balloon has been reported in series using the double-balloon or the Inoue technique.4 6 In our study, the benefit of using a large balloon was observed only in the absence of initial mitral regurgitation. This benefit was not observed in the case of mitral regurgitation before PMC, which may be related to the extent of valvular or subvalvular scarring frequently associated in such patients.
The choice of the technique for PMC remains debatable, since studies comparing different techniques have led to conflicting results,14 23 but most of these studies were not randomized, comprised a limited number of patients,23 and frequently used the Inoue balloon without the stepwise technique. The present study may contribute to the comparative evaluation of double-balloon and Inoue techniques, as standardized in a single-center series with a large number of patients, by taking into account the other predictive factors of immediate results. In the present series, there is a nonsignificant trend toward better immediate results with the Inoue stepwise technique than with the double balloon. The absence of interaction between the type of balloon and the EBDA shows that the effect of EBDA on the results does not differ according to the type of balloon used, despite differences in their shapes. The predictive model did not show interactions between the type of balloon and the characteristics of the patient. Our study therefore does not suggest a particular benefit with the use of the Inoue balloon in some subsets of patients (eg, older patients or those with severe mitral disease), as has been found in other studies.24
Validation and Testing of the Predictive Model
The following part of the statistical analysis is the validation of the model developed from the training sample. The validation on the training sample (internal validation) would have led to an overestimation of the discriminant power of the model, even with the use of cross-validation or bootstrap resampling methods.25 Validation was therefore performed prospectively on an independent test sample (external validation), which is the only way to avoid any classification bias. The good fit to the independent sample shows the robustness of the model and validates the significance and the strength of the predictors detailed above. Furthermore, the test of the model enables discriminant analysis to be performed to evaluate the application of the predictors identified as an aid to decision making for individual patients. This discriminant analysis shows that sensitivity is high, reflecting good prediction of good immediate results. Nevertheless, specificity is low, indicating insufficient prediction of inadequate immediate results. This limited discrimination does not detract from the validity of the model itself, which correctly fits data from the independent sample. The relatively low specificity revealed by the application of the model is proof of the intrinsic limitations of the prediction of immediate results of PMC, ie, the possibility of good results in patients who are at high risk for inadequate results according to the model. The possibility of good immediate results of PMC in theoretically nonsuitable cases is consistent with findings from experimental26 and clinical studies.4 5 13 27
The results of discriminant analysis may help decide whether to perform PMC in any given patient by providing a given probability of good results below which PMC should not be undertaken. Wide indications for PMC result from the choice of a relatively low threshold, for example P=.70, which allows an excellent prediction of good results at the expense of an insufficient prediction of inadequate results. Conversely, the restriction of PMC to patients who have a probability of good results ≥P=.90 is associated with a higher specificity but leads to the exclusion from PMC of many patients who would have had good immediate results. The choice of a threshold for decision making must be based on the quality of the discrimination obtained but also be adapted to each individual patient. A final decision must take into account relative efficacy and risk of alternative therapies.
Limitations of the Study
Comparison with other studies is limited by the absence of standardization of the methods used to evaluate mitral anatomy. The semiquantitative score described by Abascal et al4 has been widely used, but other publications have described different semiquantitative scores14 or qualitative classifications using an overall assessment of mitral anatomy,3 such as our classification.2 7 8 The echo scoring system used in this study has been described by the authors and is based on their experience in the selection of patients for the most adequate surgical procedure.8 The respective values of these different scoring systems cannot be evaluated, since neither anatomic validation nor comparative clinical studies have been made.5 However, all scoring systems have failed to predict the results of PMC accurately.
All procedures in the validation sample used the Inoue balloon, whereas the model was developed from the training sample comprising different types of balloon. The absence of any interaction between the type of balloon and the other variables shows that the predictive value of these variables is independent of the type of balloon used. Moreover, the type of balloon is in fact not a strong predictor of immediate results, as indicated by the adjusted odds ratio of 1.5, with a significance of only P=.09. This exclusive use of the Inoue balloon in the validation sample is therefore of limited effect, and the conclusions drawn from the validation of the model can be applied to Inoue and double-balloon techniques.
The selection of candidates for PMC must also take into account the predictors of long-term results, which are not always the same as those predicting immediate results.28 For this reason, a separate predictive analysis of immediate and late results is justified.
This large single-center series with a variety of patient subsets shows that the prediction of immediate results of PMC is multifactorial. Selection of candidates for PMC should not be based exclusively on anatomic criteria but should also include other relevant, individual characteristics; procedural variables must likewise be taken into account to predict immediate results. With regard to the practical application of these predictors for individual patients, the high sensitivity of the predictive model allows easy selection of “ideal candidates” for PMC. Conversely, the relatively low specificity demonstrates intrinsic limits in the prediction of inadequate results in patients who, according to the model, are at high risk of inadequate results. Indications for PMC should therefore not be too restrictive in such patients, especially if the clinical context is not favorable for surgery.
The predicted probability (P) of good immediate results of PMC can be calculated for any given patient: P (good results)=1/[1+exp(c)], where c is the linear function of the covariates of the logistic function: c=−4.5305+(0.6093×age)−(0.1921×previous commissurotomy)−(0.1638×MR)+(0.8331×echo group)+(0.5865×valve area)+(0.4199×type of balloon)−(0.6568×EBDA)+(1.1026×age at previous commissurotomy)+(1.0037×EBDA×MR), where MR is degree of mitral regurgitation and EBDA is effective balloon dilating area. All variables are coded as subgroups as shown in Table 5. The reference subgroup, ie, odds ratio=1, is coded 0, and subsequent subgroups are coded from 1 up.
- Received January 24, 1996.
- Revision received May 8, 1996.
- Accepted May 20, 1996.
- Copyright © 1996 by American Heart Association
Nobuyoshi M, Hamasaki N, Kimura T, Nosaka H, Yokoi H, Yasumoto H, Horiuchi H, Nakashima H, Shindo T, Mori T, Miyamoto AT, Inoue K. Indications, complications, and short-term clinical outcome of percutaneous transvenous mitral commissurotomy. Circulation.. 1989;80:782-792.
Abascal VM, Wilkins GT, O'Shea JP, Choong CY, Palacios IF, Thomas JD, Rosas E, Newell JB, Block PC. Prediction of successful outcome in 130 patients undergoing percutaneous balloon mitral valvotomy. Circulation.. 1990;82:448-456.
The NHLBI Balloon Valvuloplasty Registry Participants. Multicenter experience with balloon mitral commissurotomy. Circulation.. 1992;85:448-461.
Herrmann HC, Ramaswamy K, Isner JM, Feldman TE, Carroll JD, Pichard AD, Bashore TM, Dorros G, Massumi GA, Sundram P, Tobis JM, Feldman RC, Ramee S. Factors influencing immediate results, complications, and short-term follow-up status after Inoue balloon mitral valvotomy: a North American multicenter study. Am Heart J.. 1992;124:160-166.
Vahanian A, Acar J. Mitral valvuloplasty: the French experience. In: Topol EJ, ed. Textbook of Interventional Cardiology. 2nd ed. Philadelphia, Pa: WB Saunders; 1994:1206-1225.
Cormier B, Vahanian A, Michel PL, Starkman C, Enriquez L, Kulas A, Vitoux B, Acar J. Evaluation par e´chographie bidimensionnelle et Doppler des re´sultats de la valvuloplastie mitrale percutane´e. Arch Mal Coeur.. 1989;82:185-191.
Block PC, Palacios IF. Aortic and mitral balloon valvuloplasty: the United States experience. In: Topol EJ, ed. Textbook of Interventional Cardiology. 2nd ed. Philadelphia, Pa: WB Saunders; 1994:1189-1205.
Mickey RM, Greenland S. The impact of confounder selection criteria on effect estimation. Am J Epidemiol.. 1989;129:125-137.
Lemeshow S, Hosmer DW. A review of goodness of fit statistics for use in the development of logistic regression models. Am J Epidemiol.. 1982;115:92-106.
Tuzcu EM, Block PC, Griffin BP, Newell JB, Palacios IF. Immediate and long-term outcome of percutaneous mitral valvotomy in patients 65 years and older. Circulation.. 1992;85:963-971.
Thomas JD, Wilkins GT, Choong CYP, Chir B, Abascal VM, Palacios IF, Block PC, Weyman AE. Inaccuracy of mitral pressure half-time immediately after percutaneous mitral valvotomy: dependence on transmitral gradient and left atrial and ventricular compliance. Circulation.. 1988;78:980-993.
Cequier A, Bonan R, Serra A, Dyrda I, Crepeau J, Dethy M, Waters D. Left-to-right atrial shunting after percutaneous mitral valvuloplasty. Circulation.. 1990;81:1190-1197.
The NHLBI Balloon Valvuloplasty Registry Participants. Complications and mortality of percutaneous balloon mitral commissurotomy. Circulation.. 1992;85:2014-2024.
Feldman T, Carroll JD, Isner JM, Chisholm RJ, Holmes DR, Massumi A, Pichard AD, Herrmann HC, Stertzer SH, O'Neill WW, Dorros G, Sundram P, Bashore TM, Ramaswamy K, Jones LS, Inoue K. Effect of valve deformity on results and mitral regurgitation after Inoue balloon commissurotomy. Circulation.. 1992;85:180-187.
Reifart N, Nowak B, Baykut D, Satter P, Bussmann WD, Kaltenbach M. Experimental balloon valvuloplasty of fibrotic and calcific mitral valves. Circulation.. 1990;81:1005-1011.