Development of novel diagnostic approaches based on pulmonary physiology: Applications in acute pulmonary embolism and obstructive sleep apnea
Timon Fabius is a PhD student in the department of Research Methodology, Measurement and Data Analysis (OMD). His supervisor is prof.dr. J.A.M. van der Palen from the faculty of Behavioural, Management and Social sciences (BMS).
The most important function of the lungs is gas exchange. They are able to do so due to a sophisticated balance between ventilation (allowing inhalation of fresh oxygen rich air and exhalation of carbon dioxide rich air) and perfusion (allowing transportation of oxygen to the organs and vice versa transportation of carbon dioxide). A mismatch between the ventilation and perfusion of the lungs may cause desaturation and hypercapnia. In acute pulmonary embolism (PE) the perfusion of the lungs is affected due to an obstruction of (a part of) the pulmonary vasculature. In obstructive sleep apnea (OSA), ventilation is impaired due to a (partial) obstruction of the upper airway. Both PE and OSA share one other characteristic: in both cases novel diagnostic approaches are desirable.
As PE is a potentially lethal disease, it needs to be ruled out with a high certainty when suspected to be able to discharge the patient safely (or continue workup for some other diagnosis). To do so, a computed tomography pulmonary angiography (CTPA) needs to be performed. However, symptoms of PE (mainly dyspnea and thoracic pain) are non-specific. Consequently, the proportion of CTPA scans that confirm PE is low (approximately 25-30%). This leads to a need for novel tools to exclude PE.
OSA is often accompanied by excessive daytime sleepiness and is associated with an increased risk of cardiovascular related adverse events. The prevalence of OSA is increasing rapidly (mainly due to increasing obesity rates). However, the diagnosis of OSA requires expensive and laborious testing, which leads to a rapid increase in OSA-related health care costs and an increased load on sleep centers. Nevertheless, a substantial proportion of patients referred due to suspected OSA do not suffer from it (up to 30%). Hence, there is a need for cheap, fast, but valid tools that enable exclusion of OSA. The research presented in this thesis focused on the development of novel diagnostic tools using the consequences of ventilation / perfusion mismatches in PE (chapters 2-4) and OSA (chapters 5-7).
In chapter 2 we measured the transfer factors of the lungs for nitric oxide (TLNO) and carbon monoxide (TLCO) in subjects with suspected PE and compared the ratio of these transfer factors to the CTPA results. The transfer factor of the lungs for a certain gas depends on an alveolar-capillary membrane component (e.g. a thickened membrane will result in less diffusion) and a hemodynamic component (e.g. a decrease in pulmonary blood volume will result in less diffusion). Due to a high affinity to bind with hemoglobin, TLNO is virtually independent from the hemodynamic component whereas TLCO is approximately equally dependent from the membrane and hemodynamic component. The ratio of TLNO and TLCO should therefore provide some indication of pulmonary hemodynamics. The results of our study presented in chapter 2 showed no differences in TLNO / TLCO ratio between subjects in whom PE was confirmed and those in whom PE was excluded on CTPA. Moreover, on average, TLNO and TLCO were decreased in subjects with and without PE. This decrease was likely (at least partially) caused by a decreased alveolar volume. This decreased alveolar volume might be the result of suboptimal maximal inhalation caused by thoracic pain, which most subjects with suspected PE experience. These data indicated that the TLNO / TLCO ratio cannot be used to exclude PE.
In chapter 3 we investigated the use of volumetric capnography to exclude PE in the emergency department. Under normal physiological circumstances, the amount of carbon dioxide in exhaled air at the end of exhalation (PETCO2) is approximately equal to the amount of carbon dioxide in the arterial blood (PaCO2). PE will result in increased dead space ventilation (i.e. a part of the lungs is ventilated but not perfused). In the parts without perfusion, no carbon dioxide can diffuse into the alveolar air resulting in lowered carbon dioxide levels in exhaled air. Previous studies on capnography in PE mainly focused on the use of (a combination of) PETCO2 and PaCO2. Volumetric capnography (i.e. measurement of exhaled carbon dioxide levels as a function of the exhaled volume) enables the measurement of many more parameters than solely PETCO2. In chapter 3 we designed a novel parameter that combines several volumetric capnography characteristics that may be affected by PE. This novel parameter was defined as the amount of carbon dioxide exhaled per breath (VCO2) multiplied with the slope of the alveolar phase of the volumetric capnogram (slopeIII), divided by the respiratory rate (RR) (i.e.
). Both VCO2 and slopeIII are likely to decrease in PE whereas RR is likely to increase (to compensate for the decrease in VCO2). Thus, we hypothesized that our novel parameter is decreased in subjects with PE compared to those without. In the study presented in chapter 3 we measured volumetric capnograms in 30 subjects with suspected PE, automatically determined the novel parameter, and compared it with the CTPA scans. As hypothesized, the novel parameter was decreased in subjects with confirmed PE compared to subjects without. The area under the receiver operating characteristic (ROC) curve (AUC) of the novel parameter to exclude PE was 0.79 (95% confidence interval (CI) 0.64-0.95). A value of the novel parameter ≥ 1.90 Pa.min seemed to exclude PE with a high certainty (sensitivity 100% (95%CI 77%-100%, negative predictive value 100% (95%CI 68%-100%) and specificity 47% (95%CI 26%-69%)).
Given the small number of included subjects (and thus wide confidence intervals), validation of the novel parameter (CapNoPE) was needed. In chapter 4 we performed an external validation of CapNoPE in a dataset of an earlier study on volumetric capnography in PE. This study measured volumetric capnograms in 205 subjects with suspected PE at the emergency department. Diagnosis of PE was obtained using CTPA or proven deep venous thrombosis combined with thoracic symptoms. The results presented in chapter 4 showed that CapNoPE was again significantly decreased in subjects with PE compared to those without. The AUC of the ROC curve for CapNoPE to exclude PE was 0.71 (95%CI 0.64-0.79) and was essentially equal to the AUC of more conventional capnography parameters (with or without the need of PaCO2). Nevertheless, the diagnostic accuracy of the cutoff of ≥ 1.90 Pa.min to exclude PE was too low to use in clinical practice (sensitivity 64.7% (95%CI 52.2%-75.9%), negative predictive value 77.4% (95%CI 68.2%-84.9%) and specificity 59.9% (95%CI 51.1%-68.1%)).
In chapters 5-7 we investigated diagnostic tools for OSA. The diagnosis of OSA relies on the determination of the frequency of apneas and hypopneas during sleep, expressed as the apnea hypopnea index (AHI). According to current guidelines, a hypopnea needs to be accompanied by either an arousal or a substantial desaturation (≥3% or ≥4% depending on the definition used). A full polysomnography (PSG), in which amongst others sleep, flow, respiratory effort, desaturations and arousals can be measured, is considered the gold standard diagnostic method for OSA. However, performing full PSGs is expensive and time consuming. In an uncomplicated case a polygraphy (PG) (in which flow, respiratory effort and desaturations can be measured) is sufficient. Although a PG is substantially simplified compared to PSG it is still associated with substantial costs and warrants manual scoring. We hypothesized that the use of oximetry (possibly combined with a questionnaire) might be a further simplification that can be used to exclude OSA. To support this hypothesis we first investigated the correlation between the AHI and oxygen desaturation index (ODI, number of ≥3% desaturations per hour) in PGs (chapter 5). Specifically, we sought to identify and validate a cutoff for the ODI that could exclude OSA (defined as an AHI ≥ 5). To do so we divided 3413 PGs into a training set and validation set. In the training set an ODI < 5 seemed to best predict an AHI < 5. In the validation set this resulted in a sensitivity of 97.7% (95%CI 96.5% - 98.6%), a negative predictive value of 91.4% (95%CI 87.1% - 94.6%) and a specificity of 97.0% (95%CI 93.8% - 98.8%).
Given the high diagnostic accuracy of oximetry shown in chapter 5, we performed a prospective study on the use of automatically analyzed oximetry combined with the Philips questionnaire in 140 patients in whom their general practitioner suspected OSA (chapter 6). The Philips questionnaire consists of a combination of several OSA screening questionnaires and was developed using a middle-aged company-worker population. In the study presented in chapter 6 we investigated the diagnostic accuracy of two predefined strategies for the referral to a sleep center for OSA workup: 1) refer to a sleep center for OSA workup if the ODI is ≥ 5 and 2) refer to a sleep center for OSA workup if the ODI is ≥ 5 and/or the Philips questionnaire score is ≥ 55% (indicative of a high risk of OSA). These strategies were compared to the results of sleep center diagnostic workup. The sleep center diagnostic workup did not result in a diagnosis of OSA in 40 of the 140 included subjects (29%). The strategy to refer to a sleep center using only ODI ≥ 5 excluded OSA in 15% of the included subjects and resulted in a sensitivity of 99.0% (95%CI 94.5% - 100.0%), a negative predictive value of 95.2% (95%CI 76.2% - 99.9%), a specificity of 50.0% (95%CI 33.8% - 66.2%) and a positive predictive value of 83.2% (95%CI 75.2% - 89.4%). The two-step strategy to refer to a sleep center when the ODI is ≥ 5 and/or the Philips questionnaire score is ≥ 55%, excluded OSA in 10% of the included subjects and resulted in a sensitivity of 100.0% (95%CI 96.3% – 100.0%), a negative predictive value of 100.0% (95%CI 76.8% - 100.0%), a specificity of 35.0% (95%CI 20.6% – 51.7%) and a positive predictive value of 79.4% (95%CI 71.2% - 86.1%). Exploratively, an optimal strategy was sought to further reduce the number of referrals for OSA workup. An optimal diagnostic accuracy could be achieved if at least one of three conditions applied: 1) the Philips questionnaire score was ≥ 92%, or 2) the rounded ODI was ≥ 10 or 3) the rounded ODI was 5-10 and the Philips questionnaire score was ≥ 46.5%. This strategy would have excluded OSA in 19% of the included subjects and would result in a sensitivity of 99.0% (95%CI 94.6% - 100.0%), a negative predictive value of 96.3% (95%CI 81.0% - 99.9%), a specificity of 65.0% (95%CI 48.3% - 79.4%) and a positive predictive value of 87.6% (95%CI 80.1% - 93.1%). Concluding, the use of oximetry with or without a questionnaire enabled reliable exclusion of OSA. The advantage of the predefined two-step strategy is that oximetry would only be needed in those with a low or intermediate risk of OSA (i.e. a Philips questionnaire score < 55%). On the other hand, the advantage of the predefined strategy using oximetry alone is that a higher proportion of “unnecessary” referrals can be avoided. The explorative analysis resulted in the identification of an even higher proportion of subjects without OSA but needs to be validated before it can be used in clinical practice.
In chapter 7 we investigated whether exhaled breath analysis using an electronic nose could reliably distinguish OSA (defined as an AHI ≥ 15) from non-OSA subjects using 83 sleep center patients suspected of OSA and scheduled for a PG. We hypothesized that the many systemic (pathologic) processes caused by OSA might be reflected in exhaled breath. The results presented in chapter 7 showed that one principal component (PC4) from the exhaled breath data was significantly different in subjects in whom OSA was confirmed compared to those without OSA. However, diagnostic accuracy to distinguish OSA from non-OSA subjects using PC4 was fair at most (cross-validation value 68.7%). Nevertheless, PC4 could predict the AHI as continuous parameter (R2=0.38). The addition of known OSA-related factors in a multivariate model improved the accuracy of the prediction of the AHI (R2=0.53). We were also interested which OSA-related parameter could best predict the exhaled breath profiles. Though hypoxic burden did predict PC4 in univariate analysis (R2=0.11), in a multivariate model only the AHI added significantly to the prediction of PC4 (R2=0.43). This suggests that the frequency of breathing events rather than their duration and severity (expressed as hypoxic burden) influence processes that are reflected in exhaled breath profiles. We concluded that the exhaled breath profiles could not be used to reliably exclude OSA (when OSA is defined as an AHI ≥ 15).
In chapter 8 we discuss the results presented in chapters 2-7 and put them in context. An important topic is the definitions used for PE and OSA and the implications they may have on the data presented in this thesis..