Predictive power of extubation failure diagnosed by cough strength: a systematic review and meta-analysis

Background The predictive power of extubation failure diagnosed by cough strength varies by study. Here we summarise the diagnostic power of extubation failure tested by cough strength. Methods A comprehensive online search was performed to select potentially eligible studies that evaluated the predictive power of extubation failure tested by cough strength. A manual search was also performed to identify additional studies. Data were extracted to calculate the pooled sensitivity, specificity, positive likelihood ratio (LR), negative LR, diagnostic odds ratio (DOR), and area under the receiver operating characteristic curve (AUC) to evaluate the predictive power of extubation failure. Results A total of 34 studies involving 45 study arms were enrolled, and 7329 patients involving 8684 tests were analysed. In all, 23 study arms involving 3018 tests measured cough peak flow before extubation. The pooled extubation failure was 36.2% and 6.3% in patients with weak and strong cough assessed by cough peak flow, respectively. The pooled sensitivity, specificity, positive LR, negative LR, DOR, and AUC were 0.76 (95% confidence interval [CI]: 0.72–0.80), 0.75 (0.69–0.81), 2.89 (2.36–3.54), 0.37 (0.30–0.45), 8.91 (5.96–13.32), and 0.79 (0.75–0.82), respectively. Moreover, 22 study arms involving 5666 tests measured the semiquantitative cough strength score (SCSS) before extubation. The pooled extubation failure was 37.1% and 11.3%, respectively, in patients with weak and strong cough assessed by the SCSS. The pooled sensitivity, specificity, positive LR, negative LR, DOR, and AUC were 0.53 (95% CI: 0.41–0.64), 0.83 (0.74–0.89), 2.50 (1.93–3.25), 0.65 (0.56–0.76), 4.61 (3.03–7.01), and 0.74 (0.70–0.78), respectively. Conclusions Weak cough is associated with increased extubation failure. Cough peak flow is superior to the SCSS for predicting extubation failure. However, both show moderate power for predicting extubation failure. Supplementary Information The online version contains supplementary material available at 10.1186/s13054-021-03781-5.

application of preventive strategies (e.g. noninvasive ventilation or the use of a high-flow nasal cannula) can reduce hospital mortality [7,8]. Therefore, the key question is how to identify patients at high risk for extubation failure.
Weak cough is a predictor of extubation failure. It can be measured by cough peak flow [9][10][11][12][13][14][15][16][17]. In some studies, patients with successful extubation had a higher cough peak flow than those who experienced extubation failure [9][10][11][12][13][14][15][16]. However, another study reported that cough peak flow did not differ between patients who experienced extubation success and failure [17]. In addition, cough strength can also be measured by the semiquantitative cough strength score (SCSS) [18][19][20][21]. Given the inconsistent results found by different studies and the use of multiple methods to measure cough strength, we reviewed the literature systematically and performed a meta-analysis to assess the efficacy of diagnostic tests that use cough strength for the early detection of extubation failure.

PICO statement
P-patient: adult patients were under MV through endotracheal intubation. I-index test: cough strength was measured in all included patients. C-complement: an SBT was given to all included patients who were deemed ready to be liberated from MV. O-outcome: the efficacy of cough strength for predicting extubation failure was estimated.

Search techniques and selection criteria
This systematic review and meta-analysis was performed in conformance with the Preferred Reporting Items for Systematic Reviews and Meta-analysis statement [22]. We searched pertinent research published before June 2021 in PubMed, Web of Science, the Cochrane library, and some Chinese databases (CBM, Wanfang Data, and CNKI) without any language limitations. We also did manual searches of the reference lists of included articles to identify additional relevant articles. The studies were searched with the following key words: ("weak cough" OR "ineffective cough" OR "cough peak flow" OR "cough peak expiratory flow" OR "cough strength") and ("ventilator weaning" OR "wean from mechanical ventilation" OR "weaning from mechanical ventilation" OR "liberation from mechanical ventilation" OR "liberate from mechanical ventilation" OR "withdrawal of mechanical ventilation" OR "extubation failure" OR "postextubation failure" OR "postextubation respiratory failure" OR "reintubation").
Studies were enrolled based on the following inclusion criteria: (1) only adult patients with an endotracheal tube were involved, (2) an SBT was completed before extubation, (3) cough strength was assessed before extubation, and (4) data were available for calculating outcomes (true positive [TP], false positive [FP], false negative [FN], and true negative [TN]). The following works were excluded: (1) reviews, case reports, editorials, letters, and conference abstracts; (2) articles with no available data for patients with weak cough; and (3) articles without a definition of extubation failure. Extubation failure included reintubation, death, or the use of noninvasive ventilation due to postextubation respiratory failure.

Data extraction and evaluation of quality
All studies were independently selected by two investigators (JD and XFZ). Any discrepancies were resolved by consensus. If the researchers failed to reach a consensus, a third investigator (JPS) reviewed the data in question. The first author's name; publication year; study region; sample size; methods of assessing cough strength; cut-off value; definition of weak cough; and number of patients with TP, FP, FN, and TN were collected. If numbers of TP, FP, FN, and TN were unavailable, we communicated with the corresponding author to obtain these data. The Quality Assessment of Diagnostic Accuracy Studies 2 was used to assess the quality of the enrolled articles [23].

Statistical analysis
The data were analysed with RevMan 5.3, Meta-Disc 1.4, and Stata SE 15.0. The pooled diagnostic odds ratio (DOR), sensitivity, specificity, positive likelihood ratio (LR), negative LR, and area under the receiver operating characteristic curve (AUC) were calculated by TP, FP, FN, TN. Sensitivity = true positives/(true positives + false negatives). Specificity = true negatives/(true negatives + false positives). True positives were patients with ineffective cough who failed extubation. False negatives were patients with effective cough who failed extubation. True negatives were patients with effective cough who were successfully extubated. False positives were patients with ineffective cough who were successfully extubated. Diagnostic power was good, moderate, and poor if the AUC was more than 0.8, between 0.7 and 0.8, and less than 0.7, respectively [24]. Deeks' funnel plot was used to detect publication bias. If publication bias was present, a sensitivity analysis was performed to explore why.
Spearman's correlation coefficient is used to detect threshold effects. I 2 is used to describe heterogeneity. I 2 ≥ 50% represents significant heterogeneity. A fixed effects model was used if no heterogeneity was observed. A random effects model was selected if significant heterogeneity was observed. Possible sources of heterogeneity were explored through a meta-regression analysis.

Characteristics of the included studies
A total of 575 studies were obtained using the search strategy, and 14 studies were identified from other sources (Fig. 1). After screening titles and abstracts and reviewing full papers, we enrolled 34 studies involving 45 study arms in this meta-analysis [9][10][11][12][13][14][15][16][17][18][19][20][21]. A total of 7329 patients involving 8684 tests were analysed. The characteristics of the study arms are summarised in Table 1. A total of 23 study arms involving 3018 tests        Table 2 and Additional file 12: Text 1. Assessment of the SCSS before extubation was performed in 22 study arms involving 5666 tests. The pooled extubation failure was 37.1% and 11.3%, respectively, among patients with weak and strong cough assessed by the SCSS (Additional file 2: Figure 2). Spearman's correlation coefficient was 0.450 (p = 0.04), indicating the presence of a threshold effect. Three subgroups of studies measured the SCSS. Details are reported in Table 2 and Additional file 12: Text 1.

Quality assessment and publication bias
The quality of the included studies is summarised in Fig. 2. The main high risk of bias was the time between the removal of the endotracheal tube and extubation failure. The majority of studies judged extubation failure at a prespecified time after extubation, detailed in Table 1, except for four studies. Three study arms collected data on extubation failure during hospitalisation after extubation. And one study arm collected data on extubation failure during the ICU stay after extubation. Additional file 3: Figure 3 shows the lack of publication bias among studies that used cough peak flow to predict extubation failure (p = 0.41). Additional file 4: Figure 4 shows the presence of publication bias among studies that used the SCSS to predict extubation failure (p = 0.02). The sensitivity analysis showed that excluding Frutos-Vivar et al. 's study [34] negated the publication bias (p = 0.07). The sensitivity analysis also showed that the pooled DOR ranged from 4.08 to 5.02 and the pooled AUC ranged from 0.71 to 0.75 when one study was omitted (Additional file 5: Figure 5).

Accuracy of extubation failure diagnosed by cough peak flow
The pooled sensitivity and specificity were 0.76 (95% confidence interval [CI]: 0.72-0.80) and 0.75 (0.69-0.81), respectively (Fig. 3). Meta-regression analyses indicated that sensitivity and specificity did not vary by publication year, country, assessment of voluntary or involuntary cough peak flow, assessment of cough peak flow with an external flowmeter or a ventilator, different cut-off values, number of cases in the study arm, time to extubation failure after the removal of the endotracheal tube, or definition of extubation failure (Additional file 6: Figure 6). The pooled positive LR and negative LR were 2.89 (95% CI: 2.36-3.54) and 0.37 (0.30-0.45), respectively (Additional file 7: Figure 7). The pooled DOR was 8.91 (95% CI: 5.96-13.32; Additional file 8: Figure 8). The AUC was 0.79 (95% CI: 0.75-0.82) when cough peak flow was used to predict extubation failure (Fig. 4). The results of subgroup analyses are summarised in Table 2.

Accuracy of extubation failure diagnosed by the SCSS
The pooled sensitivity and specificity were 0.53 (95% CI: 0.41-0.64) and 0.83 (0.74-0.89), respectively (Fig. 5). Meta-regression analyses indicated that sensitivity and specificity did not vary by publication year, country, study design, method used to assess the SCSS, number of cases in the study arm, time to extubation failure after the removal of the endotracheal tube, or definition of extubation failure (Additional file 9: Figure 9). The pooled positive LR and negative LR were 2.50 (95% CI: 1.93-3.25) and 0.65 (0.56-0.76), respectively (Additional file 10: Figure 10). The pooled DOR was 4.61 (95% CI: 3.03-7.01; Additional file 11: Figure 11). The AUC was 0.74 (95% CI: 0.70-0.78) when the SCSS was used to predict extubation failure (Fig. 6). The results for cough strength assessed by the SCSS graded from 0 to 4/5, the white card test  Table 2.

Discussion
To the best of our knowledge, this is the first systematic review and meta-analysis to explore the prediction of extubation failure diagnosed by cough strength. Cough peak flow includes voluntary and involuntary peak flow and can be measured with an external flowmeter or a ventilator. The SCSS can be measured with a scale from 0 to 4/5, the WCT, or other semiquantitative scales. Both cough peak flow and the SCSS show moderate diagnostic power for predicting extubation failure. However, cough peak flow is superior to the SCSS for predicting extubation failure.
Cough strength is strongly associated with maximal inspiratory and expiratory pressure [46], which in turn can reflect respiratory muscle function. Better respiratory muscle function is associated with lower extubation failure [47]. Therefore, weaker cough strength is associated with higher extubation failure. The current study with its large sample size demonstrates that both cough peak flow and the SCSS have moderate diagnostic power for predicting extubation failure. Therefore, cough strength can be commonly used to predict extubation failure in clinical practice.
Cough peak flow includes voluntary and involuntary peak flow. Voluntary peak flow can be measured when the investigator coaches the patient to cough. Involuntary peak flow can be stimulated with an injection of 2 mL normal saline or with a suction catheter. Two studies measured both voluntary and involuntary peak flow. One showed that voluntary peak flow was better than involuntary peak flow at predicting extubation failure [11]. However, the other showed no difference between the two methods in predicting extubation failure [30]. The current meta-analysis, which enrolled 17 study arms that measured voluntary peak flow and 6 that measured involuntary peak flow, found that involuntary peak flow had much higher predictive power than voluntary peak flow. Voluntary peak flow can only be measured in cooperative patients, as it requires the Cough peak flow can be measured with an external flowmeter or a ventilator. Only one study with 126 cases measured cough peak flow using both methods [26]. And both methods showed similar predictive accuracy. However, given the small sample size in that study, its power is inadequate. Our meta-analysis, which enrolled 18 study arms that measured cough peak flow with an external flowmeter and 5 that measured it with a ventilator, found that the AUC was higher when cough peak flow was measured with an external flowmeter than a ventilator. This indicates that predictive accuracy is greater when cough peak flow is measured with an external flowmeter. However, measuring cough peak flow with an external flowmeter requires a dedicated device. This may limit the use of this method. As the AUC was 0.77 when cough peak flow was measured with a ventilator, indicating moderate accuracy for predicting extubation failure, it can be used to predict extubation failure if an external flowmeter is unavailable. However, cut-off values differ among studies. This may be related to the different devices used in the studies. Therefore, the generalisation of the measure of cough peak flow is limited by the variability in cut-off values by study, even when the method is the same.
The SCSS, which ranges from 0 to 4/5, was the most common semiquantitative method of measuring cough strength in this meta-analysis. A score of 0 indicates the weakest cough, and a score of 4/5 indicates the strongest cough [18,21]. The WCT was another semiquantitative method used to measure cough strength [13]. However, no studies compared the two methods on their predictive accuracy for extubation failure. This study found that the WCT is more accurate than an SCSS score of 0-4/5 for predicting extubation failure. The SCSS graded 0-4/5 is subjectively rated by the investigators. However, the WCT, which is scored based on the moisture on a card when the investigator coaches the patient to cough, is less likely to be influenced by the investigator's experience. Thus, the WCT can be given priority over the SCSS for predicting extubation failure.
Sensitivity was lower but specificity was higher when the SCSS (vs. cough peak flow) was used to assess cough strength. This might suggest that weak cough identified using the SCSS is actually very weak with a very low peak flow (if performed) and consequently associated with more false negatives but fewer false positives. When patients are identified as having weak cough using the SCSS, their risk of extubation failure is very high. In contrast, patients identified as having weak cough using peak flow may have a stronger cough than those identified as having weak cough using the SCSS and consequently fewer false negatives and more false positives. It may be that the SCSS is unable to detect weak cough in patients with moderately decreased peak flow (around 60 L/min).
This study has several limitations. First, the time between the removal of the endotracheal tube and extubation failure was the main high risk of quality evaluation on included studies. However, we analysed studies that defined extubation failure within and beyond 72 h. The meta-regression showed that this factor did not influence sensitivity and specificity. Second, publication bias was observed among studies that measured the SCSS. We performed a sensitivity analysis and found that the pooled DOR ranged from 4.08 to 5.02 and the pooled AUC ranged from 0.71 to 0.75. This indicates that the results were stable despite the presence of publication bias. Third, judging weak cough is difficult, as the definition of weak cough varies by study. A consensus on the definition of weak cough based on cough peak flow or the SCSS would be helpful for improving operability. Fourth, different types of SBTs were performed in the enrolled studies. The rate of successful SBTs was higher when they were performed under pressure support ventilation than under T-piece or continuous positive airway pressure [48]. However, extubation failure did not vary by type of SBT [49,50]. Therefore, type of SBT is unlikely to influence results for the association between cough strength and extubation failure.

Conclusions
Weak cough is associated with increased extubation failure. It can be assessed by cough peak flow and the SCSS. The predictive power of cough peak flow may be better than that of the SCSS for diagnosing extubation failure.