Clinical indicators of ineffective airway clearance in children with acute respiratory infection

Abstract

The identification of clinical indicators with good predictive ability allows the nurse to minimize the existing variability in clinical situations presented by the patient and to accurately identify the nursing diagnosis, which represents the true clinical condition. The purpose of this study was to analyze the accuracy of NANDA-I clinical indicators of the nursing diagnosis ineffective airway clearance (IAC) in children with acute respiratory infection. This was a prospective cohort study conducted with a group of 136 children and followed for a period of time ranging from 6 to 10 consecutive days. For data analysis, the measures of accuracy were calculated for clinical indicators, which presented statistical significance in a generalized estimated equation model. IAC was present in 91.9% of children in the first assessment. Adventitious breath sounds presented the best measure of accuracy. Ineffective cough presented a high value of sensitivity. Changes in respiratory rate, wide-eyed, diminished breath sounds, and difficulty vocalizing presented high positive predictive values. In conclusion, adventitious breath sounds showed the best predictive ability to diagnose IAC in children with respiratory acute infection.

Keywords

Clinical validation diagnostic accuracy nursing diagnosis pediatric nursing

Introduction

The establishment of good clinical indicators allows the nurse to minimize the existing variability in the clinical situations presented by the patient and to accurately identify the nursing diagnosis that represents the true clinical condition. Diagnostic accuracy requires a minimum clinical data set for determining the main focus of nursing care, establishing appropriate outcomes and selecting evidence-based interventions. Nurses usually identify a set of defining characteristics and verify a plausible relationship between the diagnostic hypotheses based on a specific clinical condition and the characteristics presented by the patient (Lopes et al., 2012).

Defining characteristics with good predictive capability allows the nurse to make assumptions related to the most probable nursing diagnosis for a given situation. This enables clinicians to make the safest inference about the presence of a specific nursing diagnosis. Studies dealing with nursing diagnoses and their components are recommended and encouraged, considering their contribution to honing the skills used by nurses in the diagnostic reasoning process (Lunney, 2009).

In the current clinical validation studies of nursing diagnoses, researchers have used a strategy for assessing the accuracy of the defining characteristics based on the approach used to determine accuracy of diagnostic tests (Beltrão et al., 2011; Sousa et al., 2013). In this case, an indicator is treated as a diagnostic test that modifies the estimate of the probability of a diagnosis being present in a given situation. Thus, the accuracy of a clinical indicator is defined as the ability of this indicator to correctly differentiate between individuals with and without a nursing diagnosis (Lopes et al., 2012).

Among the various nursing diagnoses from the NANDA International, Inc®. (NANDA-I) taxonomy, studies highlight those related to the respiratory system. The diagnosis of ineffective airway clearance (IAC) has been investigated, either alone or with other respiratory nursing diagnoses, in several populations, especially hospitalized patients (Monteiro et al., 2006; Silva et al., 2008; Yücel et al., 2011). IAC is defined as the inability to clear secretions or obstructions from the respiratory tract to maintain a clear airway (Herdman and Kamitsuru, 2014: 380). This condition is related to disease processes that contribute to increased secretions and that compromise the defense mechanisms of the airways. These processes produce changes that cause retained secretions, excessive mucus, secretions in the bronchi, and exudates in the alveoli (Di Carlo et al., 2010).

Acute respiratory infections (ARIs) stand out among the various clinical conditions associated with IAC. However, it is important to note that the prevalence of clinical indicators and measures of accuracy differ across populations (Cavalcante et al., 2010; Guirao-Goris and Duarte-Climents, 2007; Silveira et al., 2008; Sousa et al., 2013; Zeitoun et al., 2007). For example, a recent systematic review of diagnostic accuracy for clinical indicators of IAC showed that dyspnea had high sensitivity (Se) and specificity (Sp) among children with ARI but had only specificity among adults undergoing cardiac surgery (Sousa et al., 2014).

In addition, IAC presents clinical indicators that are similar to other nursing diagnoses, making the diagnostic inference difficult or even confusing because these are not mutually exclusive and one would assume it to often coexist in any particular patient. For example, the following indicators are found in other nursing diagnoses, such as ineffective breathing pattern, risk for aspiration, impaired spontaneous ventilation, and impaired gas exchange: dyspnea, cyanosis, ineffective cough, orthopnea, alteration in respiratory rate, and restlessness. Thus, this study was developed to analyze the accuracy of clinical indicators for IAC in children with ARI.

Methods

One hundred and thirty-six children with ARI were enrolled during hospitalization, and then followed post-discharge, in an open, prospective cohort study for a period of time ranging from 6 to 10 consecutive days, to verify the occurrence of IAC. The study was conducted with patients initially hospitalized in two pediatric hospitals in northeastern Brazil. Ethical approval was obtained from the institutional review board prior to initiation of the study. Parents were informed about the study and signed the terms of free and informed consent prior to data collection.

Children hospitalized for less than 48 hours and aged up to five years were included. In this study, ARI included pneumonia, bronchiolitis, sinusitis, pharyngitis, and tonsillitis diagnosed by physicians from the institution. We chose to work with a wide range of respiratory conditions because many authors recommend that diagnostic accuracy studies include individuals representing the various clinical spectrum of diagnostic interest, increasing the generalizability of the findings (Pepe, 2003; Zhou et al., 2012). Children were excluded if they had chronic diseases that changed the clinical condition of their specific ARI (e.g. congenital heart disease and cerebral palsy).

The sample size was calculated based on 95% confidence level (CI) (Z _α ² = 1.96), a conjectured Se value of the clinical indicators equal to 80% (Se = .8), and a desired width of one-half of the 95% CI of 7% (L = .07). From these values, the estimated sample size was 126 children based on the formula presented in Zhou et al. (2012)

n = \frac{[Z_{α} 2^{} Se (1 - Se)]}{L^{2}} .

However, in this study, the final sample consisted of 136 children. Because these children were evaluated for a period of between 6 and 10 consecutive days, the total number of assessments was 1128.

Measurement of variables

Data collection occurred daily over a period of 6 to 10 days, with collection occurring in clinic settings. The children were assessed using a data collection instrument that included the clinical indicators of the IAC nursing diagnosis according to the NANDA-I terminology (Herdman and Kamitsuru, 2014). This instrument was developed from literature on pulmonary evaluation (Jarvis, 2011; Potter and Perry, 2004; Swartz, 2005) and included other clinical information related to the child (gender, medical diagnosis, number of hospitalizations, date of birth, and date of admission).

Operational definitions were created for each clinical indicator to be studied, and trained members of a nursing diagnosis research group collected the data. This training lasted eight hours and included a theoretical discussion of diagnostic methods and operational definitions that would be used for data collection. Team members who perform data collection were blinded to the presence/absence of IAC. No discrepancies were found for clinical indicators because the team that collected this information strictly followed these operational definitions. Another team of raters, as described in diagnostic inference section, established the presence/absence of IAC.

Diagnostic inferences

For the process of diagnostic inference, nurses were selected from a nursing diagnoses research group. These nurses were trained to identify the IAC and then were assessed for their ability to correctly classify individuals with and without this diagnosis, through the analysis of 12 fictitious clinical cases. The aim of this strategy was to enable these nurses to achieve the same level of ability in the diagnostic inference process, as would be evident in their ability to diagnose these cases consistently and uniformly (Lopes et al., 2012). Ten nurses were divided into pairs for this step in the process, in which they would assess the data to reach a diagnosis for each patient, also known as panel diagnosis.

The use of expert panel diagnosis was described as an important strategy in studies in which no single, error-free test can be used as the gold (reference) standard (Bertens et al., 2013; Rutjes et al., 2007). The purpose of using an expert panel diagnosis as the reference standard is to provide a more accurate and reliable estimate of diagnostic accuracy for clinical indicators. Thus, health-care workers can use these diagnostic accuracy measures to guide their assessments even that does not have the same background of the evaluators in this study.

The 1128 evaluations obtained were divided into five blocks containing approximately 226 clinical cases each. Five different pairs of nurses evaluated the five blocks to determine the presence or absence of IAC. Each pair independently diagnosed the presence or absence of IAC, assessing the same children. The agreement among raters, measured by the κ coefficient, was .6861 (z = 17.45, p < .001), which is considered strong. When there was disagreement about the presence of the diagnosis, the criteria for its presence/absence were analyzed by the research team, based upon the assessments.

Statistical analysis

Statistical analysis was performed with the support of the R software, version 3.1.2 (R Core Team, 2014) using the geeM package version 0.7.1 (McDaniel and Henderson, 2014). Unadjusted generalized estimating equation (GEE) models were used to assess the association between each clinical indicator and the presence of IAC.

The GEE model was based on a structure named the autoregressive model of order 1, denoted as AR1, which assumes that the presence of each diagnostic assessment correlates with the presence of this diagnosis in the previous assessment (Van Belle et al., 2004). The indicators that were associated with nursing diagnoses, according to the GEE model, were evaluated based on the measures of accuracy.

The accuracy of clinical indicators was based on the measures of Se, Sp, predictive values (positive and negative), likelihood ratio (positive and negative), and diagnostic odds ratio (OR). The quality of clinical indicators was evaluated from the CIs for the likelihood ratio (positive and negative). In this case, a clinical indicator is considered adequate when the CIs do not contain the value 1.00.

In this research, these measures are defined below, based on the work of Lopes et al. (2012). Sensitivity represents the probability of a clinical indicator being present in patients with the diagnosis in question. Specificity (Sp) represents the probability of the absence of a clinical indicator in patients without the nursing diagnosis. The predictive value, if positive, represents the probability of the nursing diagnosis being present in patients with a specific clinical indicator. If negative, it represents the probability of the absence of the nursing diagnosis in patients without a clinical indicator. The likelihood ratio is the probability of the presence or absence of a clinical indicator in patients with the nursing diagnosis divided by the probability of this indicator in patients without the nursing diagnosis.

Results

Most of the children were male (58.1%), presented a mean age of 20.35 months (standard deviation: 3.11) and were hospitalized for a mean time of 8.29 days (standard deviation: 1.58). The most frequent medical diagnosis was pneumonia (85.3%), followed by asthma (18.4%) and pleural effusion (14.0%). A few children (11.8%) were admitted without the type of respiratory infection being specified and, in some cases, there was more than one medical diagnosis. The percentage of IAC was 93.9% on the first day and decreases during the follow-up period (90.4%, 94.8%, 81.6%, 77.2%, 72.0%, 51.5%, 41.9%, 30.9%, and 22.8%) showing a statistically significant linear trend (p < .001).

The following clinical indicators of IAC presented frequencies above 60% in the first assessment: dyspnea, change in respiratory rate, orthopnea, adventitious breath sounds, and ineffective cough. The indicator, change in respiratory rate, presented the lowest variation in the percentage values throughout the follow-up period. However, ineffective cough presented the highest frequency (range: 74.3–95.5%) (Figure 1).

Figure 1.

Clinical indicators of IAC during the research period of children with acute respiratory infection (n = 136). IAC: ineffective airway clearance.

The GEE model showed that the diagnosis of IAC was associated with the following clinical indicators: change in respiratory rate (p = .007, OR = 2.886), wide-eyed (p < . 001, OR = 68.739), adventitious breath sounds (p < .001, OR = 300.588), decreased breath sounds (p < .001, OR = 9.008), ineffective cough (p < .001, OR = 129.530), difficulty vocalizing (p = .002, OR = 10.042), and cyanosis (p < .001, OR = .035). These results are presented in Table 1.

Table 1.

Results of the unadjusted GEE model for all assessments using IAC as the response variable (yes or no) and entering individual clinical indicators as explanatory variables in the model using AR1.

Clinical indicators	OR	95% CI		p Value
Restlessness	.967	.371	2.521	.945
Cyanosis	.035	.006	.197	<.001
Dyspnea	1.056	.358	3.119	.921
Excessive sputum	.802	.322	1.999	.636
Change in respiratory rate	2.886	1.344	6.198	.007
Change in respiratory rhythm	3.222	.859	12.084	.083
Wide-eyed	68.739	1.531	3086.704	<.001
Orthopnea	2.029	.447	9.216	.359
Adventitious breath sounds	300.588	43.673	2068.869	<.001
Diminished breath Sounds	9.008	2.752	29.484	<.001
Absent cough^a	–	–	–	–
Ineffective cough	129.530	33.409	502.197	<.001
Difficulty vocalizing	10.042	2.381	42.355	.002

Note: GEE: generalized estimating equation; IAC: ineffective airway clearance; CI: confidence interval; AR1: autoregressive model of order one.

^aNo convergence of the model.

Measures of the accuracy of the clinical indicators that presented statistical significance (p < .05), based on the results obtained by the GEE model, are shown in Table 2. Adventitious breath sounds was the clinical indicator that presented the best measures of accuracy for IAC (Se: 96.62%; Sp: 81.48%; positive predictive value (PV+): 98.91%). Ineffective cough also presented high Se and predictive values. In addition, the indicators that showed high PV+ were as follows: change in respiratory rate, wide-eyed, diminished breath sounds, and difficulty vocalizing. However, the CIs for the likelihood ratio for the indicator wide-eyed did not demonstrate improvement in the probability of correctly identifying IAC (Table 2).

Table 2.

Measures of accuracy for the clinical indicators of IAC in children with acute respiratory infection.

Clinical Indicators	Se	Sp	PV+	PV−	LR+ (95% CI)	LR− (95% CI)	ROC
Cyanosis	0.56	99.16	71.43	20.96	0.66 [0.17–2.65]	1.00 [0.99–1.02]	0.498
Change in respiratory rate	59.48	55.27	83.33	26.63	1.33 [1.14–1.55]	0.73 [0.64–.084]	0.573
Wide-eyed	0.45	99.58	80.00	21.12	1.07 [0.15–7.58]	1.00 [0.99–1.01]	0.500
Adventitious breath sounds	81.48	96.62	98.91	58.12	24.14 [12.21–47.73]	0.19 [0.17–0.22]	0.890
Diminished breath Sounds	37.93	80.59	88.02	25.67	1.95 [1.50–2.55]	0.77 [0.71–0.83]	0.592
Ineffective cough	97.81	59.74	92.84	83.64	2.43 [2.00–2.95]	0.04 [0.02–0.06]	0.787
Difficulty vocalizing	11.00	97.05	93.33	22.48	3.72 [1.79–7.75]	0.92 [0.89–0.95]	0.540

Note: Se: sensitivity; Sp: specificity; PV+: positive predictive value; PV−: negative predictive value; LR+: positive likelihood ratio; LR−: negative likelihood ratio; CI: confidence interval; ROC: area under received-operator curve.

Discussion

In this study, the high prevalence of the diagnosis of IAC (91.9%) identified in the first assessment of children with ARI was similar to that found by Monteiro et al. (2006), in which this diagnosis was manifested by the entire sample. However, other studies (Silva et al., 2008; Silveira et al., 2008; Sousa et al., 2013) have shown a different prevalence of IAC, with values ranging from 31% to 66.7%. This variability may be related to several factors, such as the pathophysiology of the clinical condition of the patient, the age-group of the study population, and the sample size.

In a similar study (Silveira et al., 2008) conducted on patients with asthma, the clinical indicator adventitious breath sounds presented similar values of Se (96.43), a PV+ (84.38) and PV− (90.00), when compared with those found in the present investigation. Furthermore, the observed proportion of this clinical indicator was similar to that reported in other surveys of asthmatic children. For example, in the study by Silveira et al. (2008), the proportion was 76.2%. The studies of Chagas et al. (2011) and Mendes et al. (2012) found frequencies of 100% and 92.9%, respectively. However, in patients with cardiovascular heart disease, the adventitious breath sounds presented values below 62.5% (Fontes and Cruz, 2007; Martins and Gutiérrez, 2005).

Regarding the indicator diminished breath sounds, the children included in this study presented measures of accuracy similar to those found by Silveira et al. (2008), who obtained values of 78.57% for Se and 85.71% for PV+. In contrast, studies conducted by Sousa et al. (2013) and Silva et al. (2008), in patients with heart disease, showed that this indicator did not provide statistically significant measures of accuracy for the determination of IAC. This divergence of results emphasizes the importance of studying accuracy in patients with several clinical conditions with specific pathophysiological mechanisms.

High values in the measures of accuracy for the determination of IAC were also identified for the clinical indicator ineffective cough. However, data were not found in the literature to compare with these results. In the studies by Silva et al. (2008) and Silveira et al. (2008), the indicators absent cough and ineffective cough were analyzed together, preventing a comparison with the findings from this research.

In the present study, ineffective cough was defined as the inability to produce air movement, suddenly, noisy and intense, which tends to partially expel secretions from the airways. This indicator was evaluated by caregiver report and researcher observation, and then classified as present, absent, or not applicable. The ‘not applicable’ response referred to individuals who did not cough as well as individuals who did not have other signs and symptoms indicating secretions in the airways. Measures of the peak cough flow were not obtained due to the difficulty of obtaining this in children less than five years of age. However, the use of accessory muscles (abdominal and internal intercostal muscles) was observed during the assessment to identify an increased effort to expel secretions.

It is important to note that cough from high lung volume will not be effective when secretions are present primarily in peripheral airways, and other techniques are required to mobilize the secretions more centrally where cough clears the central airways of its secretions. In that way, even a child with normal respiratory muscle strength could have ineffective cough.

The indicator change in respiratory rate showed a high PV+ (83.33%), indicating a higher probability of the occurrence of the diagnosis of IAC in children with this clinical indicator. The high prevalence of the diagnosis of IAC in children with ARI can influence these high predictive values. The influence from the diagnostic prevalence on measures of accuracy is described in specialized literature (Pepe, 2003; Zhou et al., 2012). In the present study, change in respiratory rate was assessed with the child awake and was based on the increase or decrease in the number of breaths over a one-minute period, taking into account the patient’s age. Children with Cheyne–Stokes respiration were evaluated for two minutes to characterize the presence of bradypnea and/or tachypnea.

The indicator wide-eyed presented low frequency and nonsignificant measures of accuracy, as was similarly reported in the study of Silva et al. (2008). However, the results obtained by the GEE model showed that the presence of this clinical indicator increased the likelihood of children with ARI developing IAC.

The indicator difficulty in vocalizing presented low frequency in the present study, which may be associated with the difficulty in gathering this information because of the children’s ages. Similar studies describing results that could be compared with these data were not identified. However, Restrepo and Peters (2008) suggest that the presence of this clinical indicator in patients with respiratory infection is associated with more intense airway obstruction.

These results demonstrate that there are differences in measures of accuracy among the clinical indicators of IAC. The determination of the predictive capacity of these indicators increases the trustworthiness of the diagnostic inference process and allows the nurse to hypothesize regarding the most probable nursing diagnoses that represent the clinical situation presented by the patient.

Unfortunately, the lack of studies in the literature with a methodological design similar to the present study limited the comparison of results. Therefore, similar studies of children with ARI should be performed to facilitate comparisons. The results may have been influenced by the incorporation and diagnostic review bias. This happens when prior knowledge about the defining characteristics is incorporated during the diagnostic inference process (Zhou et al., 2012).

Although the information presented in this research contributes to accurately diagnosing IAC in children with ARI, these results should be used with caution. The children assessed were found in specialized hospitals that served patients with a higher probability of manifesting more severe clinical conditions.

Footnotes

Conflict of interest

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Beltrão

Silva

Araujo

. (2011) Clinical indicators of ineffective breathing pattern in children with congenital heart diseases. International Journal of Nursing Terminologies and Classifications 22: 4–12.

Bertens

LCM

Broekhuizen

BDL

Naaktgeboren

. (2013) Use of expert panels to define the reference standard in diagnostic research: a systematic review of published methods and reporting. PLoS Medicine 10: e1001531.

Cavalcante

JCB

Mendes

Lopes

MVO

. (2010) Indicadores clínicos de Padrão respiratório ineficaz em crianças com asma [Clinical indicators of ineffective breathing pattern in children with asthma]. Revista da Rede de Enfermagem do Nordeste 11: 66–75.

Chagas

KLM

Lima

LHO

Oliveira

EAR

. (2011) Diagnósticos de enfermagem em crianças com sinais e sintomas respiratórios: um estudo descritivo [Nursing diagnoses in children with respiratory signs and symptoms: a descriptive study]. Revista da Rede de Enfermagem do Nordeste 12: 302–308.

Di Carlo

Romano

Plano

MRA

. (2010) Children, parents and respiratory syncytial virus in Palermo, Italy: prevention is primary. Journal of Child Health Care 14: 396–407.

Fontes

CMB

Cruz

DALM

(2007) Diagnósticos de enfermagem documentados para pacientes de clínica médica [Documented nursing diagnoses for medical clinic patients]. Revista da Escola de Enfermagem da USP 41: 395–402.

Guirao-Goris

Duarte-Climents

(2007) The expert nurse profile and diagnostic content validity of sedentary lifestyle: the Spanish validation. International Journal of Nursing Terminology in Classification 18: 84–92.

Herdman

Kamitsuru

(eds) (2014) NANDA International Nursing Diagnoses: Definitions and Classification, 2015-2017. Oxford: Wiley-Blackwell.

Jarvis

(2011) Physical Examination and Health Assessment. St Louis: Elsevier Saunders.

10.

Lopes

MVO

Silva

Araujo

(2012) Methods for establishing the accuracy of clinical indicators in predicting nursing diagnoses. International Journal of Nursing Knowledge 23: 134–139.

11.

Lunney

(2009) Critical Thinking to Achieve Positive Health Outcomes: Nursing Case Studies and Analyses. Oxford: Wiley-Blackwell.

12.

Martins

Gutiérrez

MGR

(2005) Intervenções de enfermagem para o diagnóstico de enfermagem Desobstrução ineficaz de vias aéreas [Nursing interventions for the nursing diagnosis, ineffective airway clearance]. Acta Paulista de Enfermagem 18: 143–149.

13.

McDaniel

Henderson

(2014) geeM: Fit generalized estimating equations. R package version 0.7.1. Available at: http://CRAN.R-project.org/package=geeM (last accessed 23 November 2014).

14.

Mendes

Cavalcante

JCB

Lopes

MVO

. (2012) Ineffective airway clearance in children with asthma: a descriptive study. Texto & Contexto Enfermagem 21: 371–378.

15.

Monteiro

FPM

Silva

Lopes

MVO

(2006) Diagnósticos de enfermagem identificados em crianças com infecção respiratória aguda [Nursing diagnoses identified in children with acute respiratory infection]. Revista Eletrônica de Enfermagem 8: 213–221.

16.

Pepe

(2003) The Statistical Evaluation of Medical Tests for Classification and Prediction. New York: Oxford University Press.

17.

Potter

Perry

(2004) Fundamentals of Nursing. St Louis: Mosby.

18.

Core Team

(2014) R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing.

19.

Restrepo

Peters

(2008) Near-fatal asthma: recognition and management. Pulmonary Medicine 14: 13–23.

20.

Rutjes

AWS

Reitsma

Coomarasamy

. (2007) Evaluation of diagnostic tests when there is no gold standard: a review of methods. Health Technology Assessment 11: 1–51.

21.

Silva

Lopes

MVO

Araujo

. (2008) Clinical indicators of ineffective airway clearance in children with congenital heart disease. Journal Clinical Nursing 18: 729–736.

22.

Silveira

Lima

LHO

Lopes

MVO

(2008) Características definidoras dos diagnósticos de enfermagem Desobstrução Ineficaz das Vias Aéreas e Padrão Respiratório Ineficaz em crianças asmáticas [Defining characteristics of the nursing diagnoses, ineffective airway clearance and ineffective breathing pattern, in asthmatic children]. Revista da Rede de Enfermagem do Nordeste 9: 125–133.

23.

Sousa

VEC

Lopes

MVO

Araujo

. (2013). Clinical indicators of ineffective airway clearance for patients in the cardiac postoperative period. European Journal of Cardiovascular Nursing 12: 193–200.

24.

Sousa

VEC

Lopes

MVO

Silva

(2014) Systematic review and meta-analysis of the accuracy of clinical indicators for ineffective airway clearance. Journal of Advanced Nursing 71(3): 498–513.

25.

Swartz

(2005) Textbook of Physical Diagnosis: History and Examination. St Louis: Saunders.

26.

Van Belle

Fisher

Heagerty

. (2004) Biostatistics: A Methodology for the Health Sciences. New Jersey: John Wiley & Sons.

27.

Yücel

ŞC

Eşer

Güler

. (2011) Nursing diagnoses in patients having mechanical ventilation support in a respiratory intensive care unit in Turkey. International Journal of Nursing Practice 17: 502–508.

28.

Zeitoun

Barros

ALBL

Michel

JLM

. (2007) Clinical validation of the signs and symptoms and nature of the respiratory nursing diagnoses in patients under invasive mechanical ventilation. Journal of Clinical Nursing 16: 1417–1426.

29.

Zhou

Obuchowski

McClish

(2012) Statistical Methods in Diagnostic Medicine. New York: Wiley Interscience.