Abstract
Background Inspiratory muscle training (IMT) before cardiac surgery has proved to be a promising intervention to reduce postoperative pneumonia in a randomized controlled trial setting. Effects of IMT in routine care have not been reported.
Objective The purpose of this study was to investigate the effect of IMT before cardiac surgery on postoperative pneumonia in routine care at a Dutch university medical center using propensity scoring.
Design This was an observational cohort study.
Methods All candidates for cardiac surgery were preoperatively stratified by a physical therapist for low risk or high risk for postoperative pulmonary complications. Patients at high risk either engaged in an unsupervised IMT program (20 minutes a day) at home for at least 2 weeks before surgery (group 1) or received usual care (no IMT) (group 2). Results in terms of outcome measures were adjusted with propensity scores to reduce bias caused by nonrandom treatment assignment.
Results The results showed that of the 94 patients at high risk in group 1, 1 patient (1.1%) developed a postoperative pneumonia. In group 2, 8 out of the 252 patients at high risk (3.2%) developed this pulmonary complication (adjusted odds ratio=0.34, 95% confidence interval=0.04–3.38). No significant differences were found regarding median (25th–75th percentile) ventilation time (7 [5–9] hours versus 7 [5–10] hours), length of stay in the intensive care unit (23 [21–24] hours versus 23 [21–25] hours), or total postoperative length of stay (7 [6–11] days versus 7 [5–9] days).
Limitations The most important limitations of this study were confounding, incomplete data collection, and a low incidence of the primary outcome.
Conclusions Propensity scoring is believed to be a valuable tool of great potential interest to researchers in the field of observational studies. Whether IMT in routine care resulted in less postoperative pneumonia cannot be concluded.
Cardiac surgery may be followed by pulmonary complications due to the disruption of normal ventilation.1 Postoperative infection is the main noncardiac complication after cardiac surgery.2 Patients undergoing cardiac surgery often have underlying illnesses, of which several (eg, diabetes mellitus, chronic obstructive pulmonary disease) are known to be risk factors for postoperative pulmonary complications (PPCs) such as pneumonia.1,3 These complications can lead to adverse outcomes, resulting in increased hospital length of stay (LOS) or even death.1,4,5
To avoid their negative influence on postoperative outcome, PPCs need to be treated, or better, prevented. Physical therapy interventions to prevent PPCs often focus on postoperative prophylaxis using various respiratory techniques and devices.6–8 A relatively new physical therapy intervention to prevent PPCs is preoperative inspiratory muscle training (IMT). A large randomized controlled trial (RCT) showed that IMT before coronary bypass surgery significantly reduced PPCs and hospital LOS.9 Postoperative pneumonia was reduced by more than 50% in patients at high risk of developing pulmonary complications.9 Furthermore, a recent systematic review concluded that IMT significantly reduces PPCs after cardiac or upper abdominal surgery.10 These findings indicate that IMT is a very suitable preoperative tool to reduce pulmonary complications after cardiac and upper abdominal surgery and possibly also other types of invasive surgery associated with PPCs.
Unfortunately, despite promising research results, there are often large gaps between research evidence and effects in routine care.11 Randomized controlled trials often ignore the complexities of patient selection, daily care procedures, and data collection that confront health care providers in routine care. When collecting data for observational research in routine care, these complexities can lead to confounding and missing data, which makes interpretation of intervention effects in routine care challenging. The objective of the present cohort study, therefore, was to assess the effectiveness on the reduction of postoperative pneumonia of an unsupervised IMT program in routine care using propensity scoring to adjust for covariate imbalances at baseline.
Method
Background
In August 2005, at the end of the RCT by Hulzebos et al9 conducted at the University Medical Center Utrecht (UMCU), Utrecht, the Netherlands, preoperative IMT was not yet implemented as a routine preoperative intervention. In 2007, encouraged by the positive outcomes of that trial, new initiatives arose to introduce preoperative IMT as routine care at UMCU. Outcome measures were selected based on available evidence and best practice cardiothoracic procedures at that time. In contrast to the RCT, a different pulmonary risk stratification model and a different definition for pneumonia were applied. Meanwhile, a cohort study was started by a new research collaboration to investigate the effects of preoperative IMT in routine care. This article reports the results of that cohort study.
Patient Groups and Preoperative Assessment
Patients were included in the data analyses when they were accepted for elective cardiac surgery at UMCU between January 2008 and December 2009. Patients were excluded if they were admitted to a hospital before surgery or underwent an emergency procedure, cardiac transplantation, ventricular assist device implantation, or aortic dissection surgery. Depending on route of referral to a physical therapist preoperatively, patients were divided into 2 groups (Fig. 1). Group 1 consisted of patients who visited the physical therapist at the outpatient clinic (in most cases at least 2 weeks before surgery). Patients who were referred to the physical therapist through the inpatient clinic (mostly a few days before surgery) were classified as group 2.
Flow chart of included patients receiving cardiac surgery. IMT=inspiratory muscle training.
All patients were seen by a physical therapist for a preoperative assessment. During this assessment, patients received instructions and education concerning postoperative deep breathing exercises, incentive spirometry, coughing with wound support, and the importance of early postoperative mobilization. Subsequently, all patients were screened by the therapist performing the assessment for risk factors for PPCs using a validated risk model consisting of 3 factors (Tab. 1).12 This model has good predictive properties and was considered suitable for use in routine care, as it contains only 3 variables that need to be assessed preoperatively.12
Preoperative Risk Stratification for Postoperative Pulmonary Complications
The physical therapist assigned each patient to a risk stratum on the basis of available data on the 3 risk factors. Even if data on 1 or more factors were missing or unavailable, patients who had a positive score on at least 1 risk factor were considered at “high risk” of PPCs. Patients with non-missing data on all 3 risk factors, but no positive risk scores, were considered at “low risk.” When the patient displayed no positive risk scores, but data on 1 or more of the risk factors were missing, the patient was labeled as “inconclusive risk.”
If a patient from group 1 was classified as high risk, he or she received one instruction session from the same physical therapist directly after the preoperative screening and was instructed to perform IMT at home until the surgery. Patients in group 2 who were classified as high risk did not perform IMT, as there was not enough time before the surgery. Postoperatively, all patients were encouraged by the nursing staff to perform their breathing exercises (no IMT) hourly as preoperatively instructed by the physical therapist and were supported in early mobilization. All patients at high risk and inconclusive risk (in either group 1 or 2) received postoperative physical therapy as usual, which entailed deep breathing exercises, use of incentive spirometry, coughing with wound support, and forced expiratory techniques if indicated.
Inspiratory Muscle Training
Inspiratory muscle training was performed with an inspiratory threshold loading device (Threshold IMT, Respironics of New Jersey Inc, Parsippany, New Jersey). The Threshold IMT contains a calibrated spring loaded valve that provides a constant and predetermined training load during inspiration. The valve opens when the patient meets the set load during inspiration, and expiration is unimpeded. Patients at high risk (group 1) were instructed to train at home for 7 days a week, 20 minutes uninterrupted each day, for at least 2 weeks before surgery. After surgery, no IMT was performed by the patients (in either group 1 or 2). Each day, patients recorded in a diary their training duration, training intensity (cm H2O), and rate of perceived exertion (RPE)13 on a scale from 0 to 10. Initial inspiratory load was set at 30% of maximal inspiratory pressure (Pimax). Maximal inspiratory pressure at the mouth was measured at residual volume with a forceful inspiratory maneuver.14 Optimal training load was set at an RPE score of 5 or 6. Patients were instructed to increase the inspiratory load of the threshold device by 5% before the next training session if they recorded an RPE below 5 after a training session. Threshold load was unchanged with an RPE score of 5 or 6 and was reduced with an RPE score of 7 or higher. Patients performing IMT were contacted by telephone weekly after the instruction session to reinforce and evaluate the progress of their training. Training parameters such as current threshold load and possible adverse events were addressed.
Data Collection
Data concerning the preoperative risk stratification for PPCs (Tab. 1) and IMT were collected by the physical therapist performing the preoperative assessment. Data concerning preoperative patient characteristics and postoperative outcome measures were reported according to standard best practice cardiothoracic procedures at UMCU by a physician. Postoperative complications, including the primary outcome measure (ie, postoperative pneumonia), were defined by the criteria of the Society of Thoracic Surgeons.15 This definition of pneumonia is preferable in routine care because all postoperative complications after cardiac surgery are scored using this specification, resulting in international standardized data collection in the Society of Thoracic Surgeons Adult Cardiac Surgery Database.15 Pneumonia was diagnosed by a physician based on positive cultures of sputum, transtracheal fluid or bronchial washings, or clinical findings consistent with the diagnosis of pneumonia, which might include chest x-ray diagnostics of pulmonary infiltrates.15 Secondary outcome measures were ventilation time, postoperative LOS in the intensive care unit, and total LOS at UMCU. The LOS was calculated from the day of surgery onward and was recorded by a physician or secretary staff.
Statistical Analysis
Because only patients at high risk were eligible to receive preoperative IMT, those at low risk were excluded for analyses, as were patients with an inconclusive risk stratification (Fig. 1). Descriptive statistics were used to analyze the demographic variables at baseline. Summary statistics are presented as numbers with percentages in the case of categorical variables and as means with standard deviations or medians with interquartile range in the case of continuous variables. Baseline characteristics (Tab. 2) of included patients were compared between groups with the use of the Fisher exact test for categorical variables and the Wilcoxon rank sum test for continuous variables.
Baseline Characteristicsa
Training parameters of group 1 were retrospectively collected by the physical therapist from physical therapy charts. A related-samples Wilcoxon signed rank test was used to compare training load before and after IMT (group 1).
The prevalence of postoperative pneumonia and differences of secondary outcome measures in the 2 high-risk groups were compared in a logistic or Cox regression model that used propensity scores to adjust for covariate imbalances as a result of nonrandom group allocation.16 Besides group allocation, the propensity score of each patient was entered as a continuous covariate in these multivariable analyses with the primary and secondary outcome measures as dependent variables. Data were analyzed according to the principle of intention to treat. We used SPSS PASW Statistics version 20 (IBM Corporation, Armonk, New York) for all data analyses.
Propensity Scoring
Propensity scores are used to reduce selection bias by equating groups based on measured covariates and reflect for each patient the probability to end up in 1 of the 2 groups. Propensity scores are the predicted probabilities (between 0 and 1) of group allocation; a propensity score of 0.5 would mean that the patient had an equal likelihood of being classified in either group. Propensity scores were derived in a separate multivariable logistic regression model that retroactively entered all baseline variables (regardless of their level of significance) listed in Table 2 as covariates, possibly influencing group allocation and outcome.17 The propensity score includes, in a single variable, information from a large number of covariates, which is an important advantage of the propensity scoring method in contrast to classical multivariate logistic regression analysis, where several covariates are entered separately in to the model.18 The latter could easily lead to overfitting, especially when the event rate is low. In contrast to the logistic end model, overfitting is not a concern in propensity scoring models, as the propensity score is meant to be descriptive of the data at hand but not to be generalizable to other data sets.19
To calculate a propensity score for each individual, all values of the covariates need to be complete. Therefore, missing cases for the variables productive cough and decreased pulmonary function were imputed using multiple imputation after determining a patient's risk profile. Each missing case was imputed with 5 values using multiple imputation. The pooled values were used only in the calculation of the propensity score for each patient. Imputed values were not used in other analyses and were not used to determine the risk profile of a patient.
There are 3 common options to use propensity scoring for comparisons: matching, stratification, and multivariable adjustment.20 For matching, a patient is selected from the control group whose propensity score is nearest to that of a patient in the intervention group. When using stratification, patients are divided into equal-sized groups (eg, quintiles) based on their propensity scores so that comparisons are made within each stratum. The third option is to include the propensity score in a multivariable analysis of the outcome measure besides the comparison variable. In this study, multivariable adjustment was preferred because matching might result in loss of individuals for whom adequate matches cannot be found. The low number of events of the primary outcome measure would have resulted in very little events in each stratum when stratification was used.
Results
From January 2008 until December 2009, data of 950 patients undergoing cardiac surgery were collected (Fig. 1). In total, 444 patients at low risk and 160 patients with an inconclusive pulmonary risk score were excluded. Consequently, data of 346 patients at high risk were included in the analyses. Descriptive statistics are shown in Table 2. Of the patients at high risk, 94 (27%) received IMT (group 1), and 252 (73%) did not (group 2). Preoperative assessment of group 1 took place several weeks before surgery (mean [SD]=32.0 [30.4] days), whereas the assessment in group 2 was made a few days (mean [SD]=2.9 [9.0] days) before surgery.
The baseline comparison of patients in groups 1 and 2 yielded significant differences for the 3 pulmonary risk factors, median logistic EuroSCORE, and 2 surgery types, namely isolated coronary artery bypass surgery and isolated valve surgery (Tab. 2). Nine patients engaged in IMT for less than the prescribed 2 weeks due to altered surgery planning. These patients were analyzed in group 1 according to the principle of intention to treat. Mean (SD) duration of IMT was 32.0 (30.4) days.
Out of the 94 patients who were enrolled in group 1, baseline Pimax could be retrieved for 86 patients, and for 39 patients, IMT loads at baseline and after training were recorded. At baseline, median (25th–75th percentile) Pimax was 75 (50–112) cm H2O (n=86). Median (25th–75th percentile) training intensity was 20 (15–25) cm H2O at baseline and 25 (17–31) cm H2O at the end of the IMT period (n=39; P=.000). The median increase in training intensity was 25%. In 35 cases, additional notes were made on the experiences of the patients during the telephone consultations. In 23 cases, it was reported that IMT was going well. Three patients noted that the training was heavy, and 1 patient experienced dyspnea during training, but all patients continued their IMT program. Two patients needed complementary IMT instructions during the telephone consultation, 4 patients did not fully complete their IMT program due to medical (not IMT-related) or personal reasons, and 1 patient did not receive IMT instructions due to incorrect risk stratification. One adverse event concerning chest pain was reported by a patient after a long session of IMT. This patient was advised to continue IMT on an interval basis (2 sessions of 10 minutes at different moments of the day) as long as the patient did not experience chest pain.
Missing cases occurred for the values of productive cough in 32 patients at high risk (31 in group 2) and decreased pulmonary function in 69 patients at high risk (67 in group 2) (Tab. 2) and were imputed for the calculation of the propensity scores used in the adjusted regression analyses. The propensity score distribution is shown in Figure 2. Patients in group 1 (IMT) had a higher probability of being selected for that group than patients in group 2 (no IMT), with a median (25th–75th percentile) propensity score of 0.44 (0.23–0.64) versus 0.15 (0.08–0.24).
Propensity score distribution between both high-risk groups. The propensity score for group allocation is the probability given baseline variables that any patient in either group would be selected for inspiratory muscle training (IMT).
In group 1, 1 out of the 94 patients at high risk (1.1%) developed a postoperative pneumonia, whereas in group 2, 8 out of the 252 patients at high risk (3.2%) developed this pulmonary complication (Fig. 1). Without adjustment for covariate imbalances, logistic regression analysis showed an odds ratio (OR) of 0.33 (95% confidence interval [CI]=0.040–2.658). The propensity score–adjusted analyses showed an OR of 0.34 (95% CI=0.04–3.38). When only baseline variables that were significantly different between groups (Tab. 2) were entered in the propensity scoring model (instead of all baseline variables), this procedure did not lead to essential changes in the outcome of the propensity score adjusted analyses (OR=0.26, 95% CI=0.03–2.55).
Unadjusted and adjusted Cox regression analyses showed no significant differences between the 2 high-risk groups in terms of median (25th–75th percentile) ventilation time (7 [5–9] hours versus 7 [5–10] hours), LOS in the intensive care unit (23 [21–24] hours versus 23 [21–25] hours), or postoperative LOS (7 [6–11] days versus 7 [5–9] days).
A total of 160 patients (143 originally classified as group 2) were excluded from the analysis because it could not be determined whether they were at high or low risk due to incomplete data collection. Inconclusive risk stratification occurred more in group 2 because physical therapists encountered increased difficulties collecting measurements for all variables due to more inconceivable events in the inpatient clinic compared with the scheduled outpatient clinic. Of the patients excluded from group 2 (n=143), 4 developed postoperative pneumonia (Fig. 1). None of the patients with an unknown pulmonary risk score originally classified as group 1 (n=17) developed postoperative pneumonia (Fig. 1).
Discussion
In this article, we report the results of a standard of care physical therapist–instructed IMT home program on postoperative outcomes. Based on the reported results, it cannot be stated that IMT in routine care resulted in less postoperative pneumonia, decreased ventilation time, or decreased length of stay.
Previous studies showed that preoperative IMT is a promising intervention for the reduction of postoperative PPCs.9,21,22 Our data from routine care underscore the findings that IMT is simple to carry out and is well tolerated by patients. The time before surgery is spent usefully, and patients can actively contribute to the preparations for the upcoming invasive surgery. However, implementing preoperative IMT in routine care is challenging, with substantial consequences for involved staff in their daily administrative and logistics procedures. In this study, 160 patients could not be included in our data analyses due to “inconclusive risk” as a result of incomplete data collection. This result illustrates that implementation of the preoperative risk stratification, despite the pragmatic choice for the 3-factor risk model, still has certain practical shortcomings. Lack of time or absence of the patient at the appointed time in the inpatient clinic were mostly mentioned by the physical therapists as reasons for incomplete risk stratification.
The current data reflect the use and effect of preoperative physical therapy in routine care. When recording data from routine care, it is challenging to establish comparable groups with balanced covariates at baseline where covariate imbalances can lead to biased treatment effect estimates. A statistical method to improve the balance of observed covariates between groups is propensity scoring.16,23 Propensity scores can only be calculated using observed covariates. Unlike random assignment of treatments, propensity score analysis does not correct for covariates that were not assessed; therefore, the potential remains for unmeasured confounders to have influenced our findings. However, we believe that the variables included in the propensity score cover the most important covariates that could have resulted in imbalances between the 2 analyzed groups. Furthermore, Pattanayak et al23 concluded that observational study designs based on estimated propensity scores can generate approximately unbiased treatment effect estimates. Reporting effects of interventions in routine care should be encouraged to enhance insights of the results in daily care practice. Propensity scoring is believed to be a valuable tool of great potential interest to researchers in the field of observational studies and should be considered when reporting effects between nonrandomized groups with differences at baseline.
Compared with other studies reporting pneumonia, the incidence of pneumonia we found is relatively low. This low incidence of the primary outcome measure makes it more difficult to demonstrate significant results. The RCT investigating the effects of IMT before cardiac surgery reported an incidence of 6.5% versus 16.1% for pneumonia (IMT versus no IMT) in patients at high risk.9 A possible explanation for our low incidence of pneumonia is that meanwhile, in the medical field, high incidences of PPCs reported in the literature have resulted in increased attention to prevention of these complications. Increasing awareness for the use of preoperative prophylactic prescription of antibiotics, for example, possibly attributed to the decreased incidence of pneumonia.24 Furthermore, the lower incidence of pneumonia might be explained by differences in the definition and diagnostic procedures of pneumonia. According to Society of Thoracic Surgeons criteria,15 supplementary medical examination to detect pneumonia was indicated only for patients with clinical symptoms of pneumonia, whereas clinical trials, including the RCT of Hulzebos et al,9 conventionally have the medical charts and clinical records of all included patients screened by a microbiologist, possibly resulting in higher incidence rates.9,15,25 This difference in used definition together with the low incidence might imply that in the current study only more severe cases of pneumonia were detected. The role of used definitions of outcome variables on the magnitude of effect sizes is an important issue that might have influenced our results.26
There are limitations of this study that need to be reported. First, the reported data were collected by different raters, and spirometry values were obtained using different devices. We reduced the risk of interrater differences by using clear definitions in scoring demographic variables (see the definitions in the footnote of Tab. 2) and by using spirometry devices meeting the requirements of the American Thoracic Society.27 Second, due the low incidence of our primary outcome (n=10) in relation to the number of predictors (n=2) in our logistic end model, overfitting might apply to our statistical analysis for the primary outcome measure. However, there are publications that advocate relaxing the rule of 10 events per variable.28 Third, all patients in group 1 (n=94) were analyzed according to the intention-to-treat principle. Because not all patients in group 1 had the prescribed 2 weeks of training due to altered surgery planning, the reported effect of IMT may be an underestimation.
Furthermore, in our study, the adjusted OR for the primary outcome measure was less precise than the unadjusted OR, which might raise questions about the usefulness of propensity scores in this study. However, the fact that the use of propensity scores in our case resulted in broader confidence intervals does not change the fundamental choice for the use of propensity scoring. Using a more parsimonious propensity scoring model with only characteristics that were significantly different at baseline did not lead to an essential change of the outcome. When confounding might be an issue, as it was in the present study, there is the need to adjust for it. This issue of confounding, together with our low event rate, makes propensity scoring the first choice as a tool to adjust for covariate imbalances. Finally, because the aim of preoperative IMT was to reduce postoperative pneumonia, Pimax was not measured after the training period due to time constrictions. Therefore, no statements can be made regarding the effect of the intervention on Pimax in the current study. However, we did find a significant increase of 25% of the training load, and several studies have already shown that preoperative IMT improves inspiratory muscle strength and endurance.9,29,30 Because there are no reference values for the minimum increase in inspiratory muscle function, we cannot conclude that the increase of 25% in training load is clinically significant. However, compared with other studies that demonstrated increases of Pimax after IMT of 18%9 and 16%,22 the reported increase of 25% in training load appears sufficient.
In conclusion, postoperative pneumonia in our study occurred less frequently compared with other studies but is considered to be a serious adverse event. Our study showed that IMT can be part of the preoperative preparation of patients at high risk of postoperative pneumonia, but the results were inconclusive to support the evidence for IMT in reducing the incidence of this complication. Therefore, more research on its effect in routine care is recommended, as well as more research on developing a more sensitive risk stratification model to identify patients who benefit the most from IMT to increase the effect of this patient-tailored therapy. We also recommend effectiveness studies of IMT for other patient groups at increased risk for postoperative pneumonia. Examination of the cost-effectiveness and effect on other outcome measures of preoperative IMT will provide further arguments for the decision of whether to introduce IMT in routine care.
The Bottom Line
What do we already know about this topic?
Inspiratory muscle training (IMT) has been proven to increase inspiratory muscle strength and endurance in several patient groups. Furthermore, although somewhat controversial, IMT, applied in the preoperative phase, was shown to decrease the incidence of postoperative pulmonary complications (PPCs) and length of hospital stay in patients undergoing cardiac surgery.
What new information does this study offer?
This study is the first to report the effect of IMT in routine preoperative care. Propensity scoring can be a useful tool to adjust for confounding and to reduce selection bias in observational research.
If you're a patient or a caregiver, what might these findings mean for you?
Inspiratory muscle training can be applied in the preoperative phase to improve inspiratory muscle function. Due to the low incidence of PPCs after cardiac surgery in the study overall, it is difficult determine the effect of IMT on the incidence of PPCs. Therefore, if IMT is applied before cardiac surgery, it should be prescribed for those patients who are at higher risk for developing PPCs.
Footnotes
Ms Valkenet, Ms de Heer, and Dr van de Port provided concept/idea/research design. Ms Valkenet, Ms de Heer, Dr Trappenburg, and Dr van de Port provided writing. Ms Valkenet and Ms Kwant provided data collection. Ms Valkenet and Ms de Heer provided data analysis. Ms Kwant provided project management and facilities/equipment. Dr van Herwerden provided study participants. Ms de Heer, Dr Backx, Dr Trappenburg, Dr Hulzebos, and Dr van de Port provided consultation (including review of manuscript before submission).
- Received December 19, 2011.
- Accepted December 26, 2012.
- © 2013 American Physical Therapy Association
References
Issue highlights
