Normalizing daytime arterial blood gases (ABG) is a major physiological goal for patients under LTNIV (9, 10). Thus, performing daytime measurements of ABG—by puncture of the radial artery—is a standard procedure when monitoring LTNIV. Measurements are performed without NIV, at least 30 min after interrupting NIV (when possible), preferably in a sitting position. Capillary arterialized samples, taken at the earlobe after application of a vasodilating gel, are considered reliable surrogates of arterial ABG for pH and PaCO₂ (16–19). Results for PaO₂ are less reliable, with a possible underestimation by the capillary technique (20).

In stable chronic respiratory failure, pH is usually within a normal range (2, 21–24). The contribution of bicarbonates to the detection of residual nocturnal hypoventilation is debated because of the impact of a large panel of drugs and comorbidities on HCO₃ levels (25, 26). HCO₃ levels should not be used as an isolated criterion for detecting nocturnal hypoventilation, but can be indicative of nocturnal hypoventilation and should lead to nocturnal capnography if available.

Daytime ABG measurements under NIV are indicative of correction of PaCO₂ by NIV but have been shown to be a poor reflection of correction of nocturnal hypoventilation (27–29). Thus, assessing nocturnal pCO₂ is essential for identifying periods of hypoventilation related to leaks, undesired respiratory events including patient-ventilator asynchrony or inappropriate settings.

Nocturnal arterial punctures are a source of discomfort, awaken the patient, and are not recommended. Surrogates of nocturnal ABG assessment are end-tidal CO₂ (ETCO₂) and transcutaneous capnography (TcPCO₂). Reliability of ETCO₂ measurements is subject to V/Q mismatch, physiological dead space and ventilatory mode. Importantly, a normal ETCO₂ does not exclude hypercapnia, while an elevated ETCO₂ is strongly suggestive of hypercapnia but may underestimate actual PaCO₂ value (4, 30). Thus, the use of ETCO₂ as a surrogate measurement for nocturnal PaCO₂ is not recommended in most circumstances (4).

Transcutaneous capnography (TcPCO₂) is considered reliable for continuous (nocturnal) monitoring of PaCO₂ when patients are hemodynamically stable, and devices are used by an experienced team. A probe temperature of 42°C is tolerated for 8 h without any risk of skin burns in adults (31). Several studies have reviewed the performances of commercialized capnographs and show acceptable biases and limits of agreement (27, 29). A review of the recent literature on transcutaneous capnography suggested that bias values up to ± 1 kPa (7.5 mmHg), and limits of agreement up to ± 1.33 kPa (10 mmHg) were acceptable (27). PtcCO₂ however has a few drawbacks: 1/ its lag time which precludes the detection of very short events (32) and 2/ the occurrence of occasional errand values, largely dependent on the expertise of the users.

Consensual—although arbitrary—definitions for hypoventilation have been published by the American Association for Sleep Medicine (33). However, minor changes in definitions of hypoventilation can have a major impact on clinical assessment (34–36).

Pulse oximetry has the advantage of simplicity and low cost: its major drawback is the lack of specificity of desaturations, which do not allow to distinguish hypoventilation from other causes of drops in SpO₂.

Compliance is directly related to efficacy of NIV (on ABG and symptom control), and prognosis; it may also reflect patient discomfort, which impairs health related quality of life. Analysis of use of LTNIV (compliance) has 3 components: total time spent with the ventilator, pattern of use and evolution of these parameters (Figure 1). Total time spent under NIV has to be long enough for NIV to have a beneficial impact on daytime and nocturnal ABG, and, in some cases, on total energy expenditure (37, 38). It is usually considered that using NIV <3:30 h per 24 h is insufficient and does not allow resetting of respiratory centers and resting of respiratory muscles. Borel et al. described a “U-shaped curve” relationship between compliance and prognosis (39). Too low values provide insufficient benefit, while higher values indicate more severe respiratory impairment (39). In COPD, a meta-analysis by Struik et al. suggests that a daily use of NIV for at least 5 h is required to improve PaCO₂ (40). In neuromuscular disorders such as amyotrophic lateral sclerosis or myotonic dystrophy, there is a clear relation between adherence to treatment and survival (41–44). Thus, the best compromise between patient comfort and efficacy must be sought for. Pattern of use (fragmentation vs. continuous use) can be an indicator of discomfort and side-effects (leaks, pain or discomfort related to interface), co-morbidities (nycturia, use of diuretics) or symptomatic patient-ventilator asynchrony (Figure 1) (7, 45, 46). Evolution of total time spent under NIV and changes in pattern of use may be indicative of an exacerbation with a risk of hospital admission (47).

Leaks in LTNIV are either intentional (i.e., related to exhalation valves to avoid rebreathing) or unintentional (i.e., caused for instance by a poorly adapted interface, too high insufflation pressures, or mouth leaks). Intentional leaks depend on the type of mask and insufflation pressures: interfaces have their predefined leak vs. pressure curves (48). In barometric modes, which are presently by far the most frequently used, ventilators adapt the airflow generated by the turbine to the intentional leak to ensure that the pressure attained in the airways reflects settings determined by the clinicians. Unintentional leaks are by definition unpredictable, most often variable, may differ between inspiratory and expiratory phases, and can be influenced by elements such as facial morphology, type of interface (facial vs. nasal mask), body position, sleep stages or ventilator settings. More importantly, they may interfere with the detection of patient's changes in inspiratory flow (which triggers pressurization or induces cycling). Leaks are thus the most frequent etiology of patient-ventilator asynchrony. They may induce significant and sometimes severe desaturations due to hypoventilation or to a “dilution effect” in patients with supplemental oxygen and can impact on quality of sleep (49). Also, leaks have a detrimental effect on reliability of data provided by ventilator software such as tidal volume, minute ventilation, or percentage of cycles triggered or cycled by the ventilator (3). This may be particularly problematic in automated modes with volume targeting or providing a continuous assessment of airway resistance.

Most built-in monitoring systems coupled to LTNIV ventilators provide data about leaks. However, leaks are estimated and reported differently according to manufacturers, which may misguide the clinician (i.e., average leaks, average leak only during expiratory phase, average leaks with or without intentional leaks) (3, 50, 51). Bench test studies have shown that reliability of reported leaks may vary considerably according to device used and to absolute value of leaks (3, 5, 52). Arbitrary threshold values for acceptable leaks are commonly reported by several manufacturers: the threshold of 24 L/min often referred to was first mentioned in a study by Teschler et al. (53): the relevance of this value is questionable today since the efficacy of ventilator turbines and their capacity to compensate for leaks has improved substantially. A statement by the French GAV-O₂ group suggested that leaks are considered troublesome essentially 1/ if they impact on patient comfort and compliance and 2/ if they cause recurrent nocturnal desaturations and episodes of hypercapnia (46).

For the patient, the variability of leaks and peak levels even of short duration may be a more important source of discomfort than the mean or median leak level.

Tidal volume (V_T) and minute ventilation (MV) are important goals for the clinician when determining ventilator settings: a usual target for V_T is 8–10 ml/kg of ideal body weight. The accuracy of V_T, and thus MV, depends on the reliability of ventilator software, on pressure values and on leaks (54, 55). In a bench study of 7 home ventilators, bias between measured and reported V_T ranged from 66 to 236 ml (3). This may be problematic in automated volume-targeted modes. A high variability of tidal volume may per se be suggestive of leaks.

Quantifying the percentage of cycles triggered by the patient is theoretically an important information. It allows to determine to what extent the patient's respiratory rate (RR) is captured by the ventilator, i.e., to what extent ventilatory support approaches a controlled mode (Figure 2). This may be a goal for clinicians in neuromuscular disorders for instance, to rest inspiratory muscles, or to decrease residual apneas in obesity hypoventilation (56). However, in the presence of leaks, and/or upper airway obstruction, a low percentage of triggered cycles may reflect the fact that the ventilator does not detect the patient's inspiratory efforts. Conversely, a high percentage of cycles triggered by the patient, i.e., a spontaneous mode, may be an indicator of patient-ventilatory synchronization. This parameter must therefore be interpreted with caution, especially in the presence of leaks.

Ventilators also provide the percentage of respiratory cycles cycled by the patient i.e., for which the predefined percentage of peak inspiratory flow is reached. A high percentage of respiratory cycles cycled by the patient may be considered as evidence of appropriate patient ventilator synchronization. Conversely, a high percentage of cycling by the device may reflect either a “controlled mode” or leaks with or without patient ventilator asynchrony.

The presence of residual respiratory events is frequent and has a documented impact on quality of sleep and sleep structure (23, 57–60) (Figures 3–5). Severe patient-ventilator asynchrony may compromise the efficacy of NIV. Also, events such as recurrent upper airway obstruction may affect survival in ALS (9). It is thus important to evaluate, and correct these events.

Most devices report residual respiratory events occurring under NIV: they are labeled as apnea, hypopnea, and sometimes even classified as “central” or “obstructive” according to complex algorithms and estimation of upper airway resistance using the force oscillation technique, the shape of the flow curves or both. The relevance of these parameters in LTNIV is not well-established: these parameters are derived mainly from the use of CPAP in sleep-related disordered breathing (SRDB), and do not take into account the higher level of complexity of undesired residual (or de novo) respiratory events associated with LTNIV. Very few studies have assessed the reliability of these data. Georges et al. showed that the apnea-hypopnea index (AHI) reported by one specific device in patients with obesity-hypoventilation was well-correlated with a manually read polysomnographic assessment, and that a threshold value of 10/hour for AHI allowed to discriminate with a high sensitivity and specificity subjects who were correctly ventilated vs. those for whom adjustments were required (61). In a similar study, Alvarez et al. compared manually scored polygraphy with built-in software data in 26 patients with obesity-hypoventilation: automated analysis of residual events was well-correlated with polygraphy; however, manually scored tracings from ventilator software provided more consistent results than automated software (62). More recently Aarrestad et al. compared manually scored polygraphy and data from ventilator software in 67 subjects on LTNIV for restrictive disorders: a value of AHI above 7.2/hour reported by ventilator software had a sensitivity of 93% (95%CI: 68–100) and a specificity of 92% (95%CI: 81–98) for detecting patients with an AHI > 10/hour scored by polygraphy (34).

Ventilator software provides the possibility of a detailed cycle-by-cycle analysis of pressure and flow curves, while simultaneously providing curves of estimated leaks and tidal volume. Oximetry can be added to most devices to complete this analysis. This allows the detection of residual obstructive events with or without persistence of respiratory drive (63), may show periodic breathing, and patient ventilator asynchronies (rate asynchrony such as multiple triggering, auto-triggering, ineffective efforts, or intracycle asynchronies such as underassistance, or overshoot) (49). Coupling the detailed analysis of ventilator curves with oximetry and leaks allows a better assessment of relevance and mechanisms involved, and provides a rationale to adapt ventilator settings. Illustrative cases of contribution of pressure and flow curves are shown in Figures 2–5.

Respiratory polygraphy (PG) or polysomnography (PSG), preferably coupled with TcPCO₂, are considered as gold-standard for documenting undesired respiratory events occurring under LTNIV. These events have been reported by several studies (34, 45, 57, 58, 64, 65) and defined in detail by the SomnoNIV group (63). However, PG or PSG under NIV requires expertise. PSG is not easily available in many countries, and thus use of PSG for titrating and/or monitoring LTNIV, although recommended by the AASM (American Association of Sleep Medicine), is not standard practice (66). Some specialized centers use PSG for titration and follow-up (67). Proposed algorithms recommend PSG or PG only after a thorough analysis of symptoms, PtcCO₂, ventilator software and correction of leaks (4, 68) (Figure 6).

Specific HRQL questionnaires have been designed for patients under LTNIV, such as the SRI (Severe Respiratory Insufficiency Questionnaire) which is the most widely used in this setting, with 16 translations and an “App” version available (69, 70). This type of questionnaire is however mostly designed for clinical studies, and not for follow-up of individuals.

A disease-specific symptom score (S³-NIV) has been recently developed (71) and is undergoing several translations: it is an 11-item Lickert scale type questionnaire which can be auto-administered, and provides 2 sub-scores: “respiratory symptoms” and “sleep and NIV-related side effects.” This score is perfectly suitable for clinical practice. It can be coupled with simple tools for assessing mood disturbances such as the Hospital Anxiety and Depression scale (72). These tools do not replace however the necessary face-to-face exchange with the patient and/or his/her caregivers to appreciate the relevance of findings downloaded from ventilator software and mentioned in symptom scores such as the S3-NIV.

NIV should be systematically monitored (4, 46). As mentioned previously, residual nocturnal hypoventilation, unintentional leaks, patient-ventilator asynchrony or abnormal respiratory events are frequent under NIV and have a negative impact on patient-related outcomes such as symptoms, HRQL and survival. It has been estimated that approximately one third of ventilated patients with normal daytime ABG and nocturnal SpO₂ had residual nocturnal hypoventilation under NIV (28, 35, 73). Withholding from performing regular testing of NIV efficacy could therefore be detrimental.

Optimal modalities for monitoring of long-term ventilated patients remain a matter of debate. As previously mentioned, complete polysomnography (PSG) under NIV is performed by some groups, but is not feasible in most centers on a routine basis (74).

Alternative tools (such as oximetry, TcPCO₂, or ventilator software) can be used alone or in combination. A step-by-step strategy starting by ABG and nocturnal SpO₂ has been proposed by the SomnoNIV group (4) (Figure 6).

Over the past years, the use of TcPCO₂ has been simplified (less frequent changes of probes, improved software and estimation of drift). Failure to retrieve data is rare (75) and instrumental drift of TcPCO₂ is a minor problem when used by an experienced team (27). Several studies have shown that continuous TcPCO₂ recording provides an accurate picture of the overnight time course of PaCO₂ in CHRF under NIV. Experts propose different thresholds to define significant nocturnal hypercapnia: maximal TcPCO₂ >49 mmHg, TcPCO₂ >49 mmHg for >10% of recording time, TcPCO₂ >55 mmHg for ≥10 min or an increase in TcPCO₂ ≥10 mmHg above awake supine value to a value exceeding 50 mmHg for ≥10 min (33). The choice of a clinically relevant threshold may be influenced by 1/the bias between arterial and transcutaneous PCO₂ according to the device used, 2/the etiology of the underlying chronic respiratory failure and 3/PaCO₂ levels when NIV is started. Beyond this debate, capnography is currently a well-accepted surrogate procedure for diagnosing nocturnal residual hypoventilation in ventilated patients. Simultaneous recording of SpO₂ with the same ear probe improves the clinical contribution of nocturnal transcutaneous capnography. Indeed, sampling rate and averaging of SpO₂ and TcPCO₂ recordings are different. Therefore, SpO₂ traces can detect brief desaturations related to short ventilatory events while TcPCO₂ traces allow evaluating overnight trends because of a longer lag time (4).

Significant leaks and/or abnormal residual respiratory events (i.e., flow reduction or patient-ventilator asynchronism) are frequently detected in patients with normal nocturnal TcPCO₂ and SpO₂ (23, 59, 64).

In a study comparing different tools and their combination, Georges et al. demonstrated that combining the signals provided by TcPCO₂ and data from ventilator software provides the best noninvasive assessment of NIV efficacy (73). This approach allows monitoring of LTNIV at the hospital or at home without complex logistics. Interpretation of the results is simple and further analysis of detailed raw data provided by ventilator software can help clarify the underlying mechanism to optimize NIV settings, thus limiting the use of PSG to more complex cases.

A suggested strategy for nocturnal NIV monitoring and parameter optimization is summarized in Figure 6.

Several studies have reported the—sometimes frequent—occurrence of patient ventilator asynchrony (PVA) in patients under LTNIV (23, 57–59, 63, 64, 76) (Figures 4, 5). Their specific semiology and a suggested framework for their analysis has been proposed by the SomnoNIV group (49). Briefly, after correction of leaks and obstructions, which are the major contributors to PVA, a visual analysis of tracings is necessary to detect the presence of rate asynchrony and/or intracycle asynchrony. Rate asynchrony is further subdivided into multiple triggering, uncoupling and ineffective efforts, while intracycle asynchrony can reveal either flow asynchrony (under-assistance) or phase asynchrony (premature or delayed cycling). Clinical relevance of PVA is still a matter of debate, especially if not associated with discomfort and hypoventilation and/or nocturnal desaturation. Scoring of PVA is not yet standardized in the medical literature, and prevalence can vary considerably according to working definitions (23). Seeking for PVA seems however appropriate after correction of leaks in the presence of symptoms and/or unexplained desaturations.

Although the use of tele-monitoring has become in many countries standard practice for following patients treated by CPAP and Sleep-related breathing disorders (SRBD), and, in some countries, is even required for its reimbursement, this is not the case for LTNIV. Many devices used for LTNIV have the necessary technology built-in, but each manufacturer has its software, and way of reporting certain parameters, representing a challenge for clinicians. Reliability of data remains an issue for some data, as previously discussed (77). In spite of these limitations, home titration and follow-up of LTNIV are emerging strategies. Use of tele-monitoring as an adjunct to home initiation of LTNIV is also an option, and has been shown to be cost-effective in selected indications such as CHRF related to neuromuscular disorders or other restrictive lung diseases (78, 79). Several studies suggest a benefit on healthcare utilization (emergency visits, hospital admissions) and cost-effectiveness of tele-monitoring in ALS (80–82). Preliminary studies have explored the possibility of early detection of exacerbations of COPD under LTNIV (47). For instance, changes in number of hours of NIV use may be a sign of imminent exacerbation, of worsening of underlying disease, or of discomfort, and can be a useful signal for early home intervention (7, 47). On-going trials (reviewed by Borel et al.) (77) will help clarify the modalities of a cost-effective use of tele-monitoring in NIV, and importantly, its benefit (or absence of) on relevant outcomes (such as HRQL, survival, hospital admissions). Points of potential interest for telemonitoring are: follow-up after initiation of NIV following an acute episode of respiratory failure (effective daily use, leaks etc.), progressive titration of NIV after home initiation of LTNIV, and, once specific algorithms will have been validated, early detection of exacerbations or deterioration of disorder causing CRF. The present tools for tele-monitoring already provide a limited “traffic light system” (red, yellow and green icons) usually based on leaks, AHI and compliance. A more elaborate signaling system, using limits for tidal volume, minute ventilation, respiratory rate etc. has yet to be validated in clinical practice, the risk being an unwarranted increase of “red lights.” The possibility of adapting certain parameters at a distance (pressure support, expiratory positive airway pressure, or back-up respiratory rate) allows progressive changes in these settings for increased patient comfort and tolerance, and reduces the requirement for in-hospital titration (83).

The concept of “Big data” refers to access to a massive quantity of data, its speed of acquisition, and the diversity and heterogeneity of data sources, with potential uncertainties regarding data quality (84). The widespread use of telemonitoring for CPAP has opened the way to a new approach for exploring patient pathways, identifying potentially relevant phenotypes, and elements impacting on adherence and efficacy. For instance, Liu et al. provided very useful data for a better understanding of trajectories of patients with central sleep apnea (emergent, transient, persistent) after initiation of CPAP (85). Technical information such as the impact of changes of interface can be easily assessed through daily longitudinal follow-up data at a large scale. Detailed analysis of day by day use of devices reveals behavioral patterns which improve our understanding of how patients adhere and accept their treatment (86, 87). Recently, it has been possible through telemonitoring to evaluate the impact of the lockdown during the COVID pandemic on adherence to CPAP or NIV (88, 89). Undoubtedly, in a near future, this type of information will become increasingly available and provide a useful adjunct to clinical trials.