Since oxygen therapy can increase PaCO2, as explained in Sect. 2.1, it is beneficial to monitor PaCO2 not only during acute oxygenation therapy but also possibly during LTOT (especially during sleep ) (see Sect. 2.1). Mortality in acute exacerbations of COPD was reported to be associated even with an increase in chronic stable levels of PaCO2 , which shows that it is also beneficial to regularly monitor stable (baseline) CO2 levels.
The standard for obtaining PaCO2 is ABG, which is an invasive procedure with risks, but also painful, time consuming, and discrete. Even though not perfectly following the PaCO2 obtained by ABG, transcutaneous monitoring has its advantages: it is painless, noninvasive, and continuous, with almost no sleep disturbance, and possibly self-manageable. It is therefore possible to use PtcCO2, along with pulse oximetry, instead of ABG, for titrating any oxygen therapy. Moreover, since PtcCO2 is continuous, the titration can be without delay.
PtcCO2 is not measured routinely in remote settings due to the lack of inexpensive PtcCO2 monitors. Even though it is likely that the prices of PtcCO2 monitors will reduce over time, as happens with all electronic devices, an open question is how to compensate for the currently high prices. One possibility is for PtcCO2 monitors to be used by medical personnel that periodically visit multiple COPD patients for the purpose of monitoring the baseline PaCO2 regularly, and/or in the minutes after the onset of oxygen therapy. This entails a need for a new software system that will keep track of each patient’s measurements and integrate them with electronic health records. Another possibility is for the same device to be shared between multiple patients for periodic use during 1 day and night. This would be enough to access how the PaCO2 changes during the LTOT, which is particularly important during the night, as explained in Sect. 2.1.
One current weakness associated with transcutaneous measurements is the approximate 2-min lag time for the PaCO2 changes to be reflected in the PtcCO2 . An open scientific question is could the changes in PaCO2 be predicted to compensate for the lag?
It has been suggested that patients who are known to be at risk of hypercapnic respiratory failure should be given oxygen “alert cards” and should be instructed to show these cards to medical personnel [24, 47]. The purpose of the cards is to warn medical personnel that the patients should be given controlled rather than high-concentration oxygen therapy and to adjust the therapy based on the previous ABG results stored on the card because hypercapnic respiratory failure can occur even if the targeted SaO2 is below 88%. It is important to investigate how this digital alert card can be employed in a convenient form, like, for example, as a smartphone application, which could be accessible to medical personnel even if the patient is not capable of presenting the alert card on his/her own.
ECG, ECG-derived RR, and HR
Reports show that the primary cause of death for COPD patients is cardiac failure [64, 65]. Myocardial infarction is the co-morbidity with the greatest potential for treatment and prevention to improve the prognosis of COPD patients . In general, cardiovascular diseases are the most frequent and important diseases coexisting with COPD (see Sect. 2.2.) It is therefore of the greatest importance to continuously monitor cardiac electrical activity for patients at risk, and issue alarms when dangerous event are detected.
Recent developments in ECG technology have resulted in PECGs that possess convenient features for remote monitoring (see Sect. 3.2). They are small and wireless. Patients can place them on their own, freely sleep with them, and even take a shower.
They are most often single lead devices. One lead is enough to provide HR  which can be used to detect arrhythmias and cardiac arrest. Some arrhythmias need to be terminated instantaneously (see Sect. 5.4); therefore, it is crucial in that situation for a system of remote monitoring to alert people in the vicinity of the patient and direct them to the nearest defibrillator (which are becoming more and more present in the environment). PECGs can also be used to obtain specific indicators, like heart rate corrected QT interval (see Sect. 5.4), relevant for COPD patients.
It has also been shown that reliable RR can be estimated from ECGs produced by them , which eliminates the need for separate respiratory sensors. This is of additional significance for COPD patients, since RR is one of the major parameters for detecting COPD exacerbations (see Sect. 4.7).
Figure 5 shows how RR can be extracted from amplitude-modulated ECG signal obtained using a PECG. For details of the algorithm used, see . Respiration modulates the ECG also in terms of HR (respiratory sinus arrhythmia (RSA)). Even though it has been shown that extraction from HR provides a good approximation of the mean RR if the time series is longer than 1 min for young supine subjects, it is significantly less accurate for elderly people . There are also other approaches for extracting the RR from ECG. For a recent evaluation of four different methods, i.e., filtering, R and RS amplitude, and QRS areas, please refer to .
Since they are still not a part of standard clinical care (and consequently not shared among patients) and are not covered by health insurance, the disadvantage of the PECGs is that they have high cumulative consumer costs. Moreover, the customers can be dependent on the device company for raw and aggregated data retrial.
The crucial current deficiency of PECG sensors is the data-processing time, which can be after the recording (ZIO XT) or after the transmission to the company’s data networks (SEEQ), where the processing time depends on the processing abilities of the company’s data center. The Savvy sensor, however, comes with stand-alone computer software for a basic analysis of the acquired ECG that a customer can use on his/her own. The data analysis algorithms that can run directly on the PECG sensor have just recently started to be developed, an example being the HeartSaver reported in .
Since acute life-threatening events are the most important, it is obvious that there is a need for further developments to enable wireless monitors that can not only record ECG to estimate the arrhythmia burden but more importantly detect life-threatening events in real time and send an alarm to the relevant stakeholders for a timely diagnosis and treatment.
Even though PECG sensors detect more reportable events than Holter monitors, largely because of the significantly longer recording times, there is evidence that Holter monitors detect more events in the same period of time . This might be because Holters are multichannel devices, but also due to the differences in detection algorithms. Consequently, for the PECG sensors to become a standard diagnostic tool in clinical practice, it is essential for future studies to specify the abilities of each PECG sensor and the associated detection algorithms when detecting each type of arrhythmia and conduction system disease.
The PECG sensors can potentially be used even for a 12-lead ECG synthesis [98, 110, 111]. Their additional advantage is that they can be used to obtain specific ECG leads for specific purposes [112,113,114].
Because of an autonomic nervous system dysfunction, COPD patients have been shown to exhibit decreased HR variability compared to control subjects, as well as reduced levels of all the linear exponents and a decreased short-term fractal exponent of the intervals between beats . The power of the HR-related features for predicting exacerbations remains to be investigated.
Empowering patients and personalization
How can we make the system for the remote monitoring of COPD patients easy to use and able to adapt to each patient’s specific needs? A recent systematic review  shows that patients were “generally satisfied and found the systems useful to help them manage their disease and improve healthcare provision.” The review, however, indicates a number of usability problems that need to be overcome in future research. These include the short battery life, a confusing power supply, and the need to provide real-time feedback. Based on previous research by Botsis and Hartvigsen , the authors also emphasize that the security and confidentiality of the collected data should be satisfied. Interestingly, the authors also state that the lower compliance is related to the frequency and timing of the data collection and transmission, which is carried out at discrete time points during the day and requires effort from the patients. It is to be expected that the automatic collection and transmission of data will increase the patients’ compliance with the remote monitoring systems.
According to the study made by Gravil et al. , 80% of patients would be happy to be treated at home for uncomplicated exacerbations instead of being admitted to a hospital. In their study, however, the patients were visited by nurses who monitored adherence to the recommended treatment and offered reassurance, support, and education. The only measurements used were spirometry and SpO2, which were applied by the nurses. If the patients were to use a remote monitoring system on their own, we would expect the acceptance of the home treatment to be significantly lower. One of the reasons for this could be technophobia, which according to the American Telemedicine Association  can be reduced among users by tailoring systems to the specific needs of each user population. Furthermore, there is evidence that the promotion of easy-to-use systems and more training sessions, to make patients more familiar with the system, should improve the acceptance of these remote monitoring systems [20, 118]. Contributing to the acceptance should also be the provision of additional security that comes from the fact that the patients’ symptoms will be recognized early by the remote monitoring system and that they will be contacted and treated if any deterioration occurs.
One important goal of any COPD monitoring system should be to enable patients to better understand the disease, to familiarize themselves with symptoms and ways to control them, and to empower the patients to be more involved in their healthcare, so that they can recognize the exacerbations at an early stage.
Monitoring and controlling physical activity, exercise, and reference test
Patients with COPD have significantly lower levels of physical activity (PA) than healthy controls , and even lower than people with some other chronic conditions . Even though PA is recommended for COPD patients (Sect. 2.3), most patients do not follow these recommendations . Remote monitoring may have a positive motivating effect on COPD patients to increase PA [121, 122] by providing counseling and feedback, but it is prudent to monitor COPD patients’ physiological parameters (PaCO2, PaO2, ECG) during PA also because of the health risks involved (see Sect. 2.3). For the remote assessment of PA, both questionnaires and motion sensors, like step counters and accelerometers, can be used, but a recent investigation showed that questionnaires provide overestimates of the true PA .
Questionnaires are mainly used for research purposes, especially in epidemiological studies. There have been more than 15 different questionnaires developed for COPD patients . From the perspective of continuous monitoring, the questionnaires can be implemented in an application for PCDs and be used by the patients on an everyday basis (see Sect. 4.5).
Pedometers are devices that count the number of steps. From the number of steps counted, it is possible to estimate not only the distance traveled but also the energy expended. However, these estimates lack precision, especially when the walking is at slow speed, which is typical in patients with COPD . Despite being imprecise, there is evidence that pedometers are motivating for patients to increase and maintain their levels of PA when used alone , or together with a PCD application providing individualized activity goals and allowing occasional telephone contacts with caregivers .
Accelerometers are devices that measure acceleration. The measurements obtained with accelerometers reflect body movement. For estimating PA levels and energy expenditure, they can be combined with pedometers and physiological sensors, e.g., HR and skin temperature, to provide valid estimates of PA levels [67, 125].
It is useful to know that levels of PA cannot be accurately predicted from resting lung-function parameters, i.e., spirometry .
There are also reference tests for evaluating the progression of the disease. One of them is the “6-min walk test” during which the patient walks the longest distance he/she can, while his/her blood saturation is monitored with a pulse oximeter to assess exercise-induced oxygen desaturation which has an additional prognostic values besides the 6-min walk distance . There is evidence of a positive association between the 6-min walking distance and the PA .
A questionnaire implemented on a PCD can be provided to the patient to be filled in every day. They can include fields like how much coughing, how much sputum production, how much breathless—dyspnea, general feelings, PA assessment, and other questions that can be found in existing questionnaires.
The most comprehensive COPD-related questionnaires are the Chronic Respiratory Questionnaire (CRQ)  and the St. George’s Respiratory Questionnaire (SGRQ)  (online ). The SGRQ’s score has been associated with anxiety and depression, two major comorbidities in COPD.
The latest GOLD executive summary  considers CRQ and SGRQ as being “too complex to use in clinical practice” and promotes shorter measures, like the COPD Assessment test, as more suitable. Still, the questionnaires used in the research of the remote monitoring of COPD are diverse. Some of the researchers have even developed their own questionnaires .
A drawback of self-reporting is that it can be difficult for patients. There is evidence that only a minority of patients can “log discrete episodes of increased breathlessness, cough and purulent sputum” . This is one of the reasons for preferring the continuous automatic monitoring of physiological parameters instead of self-reporting.
Monitoring medication application and adherence
An appropriate medicament therapy can reduce COPD symptoms, reduce the frequency and severity of exacerbations, and improve health status and exercise capacity, but “existing medications for COPD have not been shown conclusively to modify the long-term decline in lung function” . In particular, exacerbations are treated with bronchodilators, systemic corticosteroids, and antibiotics, and there are new drugs developed all the time . The corticosteroids can cause hypertension, which calls for blood pressure measurements and ECG monitoring during corticosteroid therapy. Stopping regular medication, such as diuretics and/or bronchodilators, on the patient’s own initiative might increase symptoms .
It has been suggested for more than 10 years now that remote monitoring should be used as the “gold standard” for medication adherence measurements , but how to effectively monitor medication compliance and how to motivate patients to take their medicaments are challenges that still remain. PCD applications that provide reminders and feedback to patients could be useful as motivators. As for monitoring compliance, intelligent packaging for medicaments, equipped with electronics for collecting and transmitting the data about usage, can provide controlled access to medicaments in terms of keeping track of how many medicaments have been used and at which times. This enables monitoring of the adherence to the therapy, the evaluation of patterns of medicament use, and monitoring the dose-response relationship . Obviously, smart packaging can be tricked into tracking adherence since it is not able to detect whether the pills are actually swallowed.
Perhaps the most impressive recent development in monitoring medication adherence, which cannot be easily tricked, is the Proteus Digital Health ingestible sensor that measures medication ingestion and adherence patterns in real time . It is a system of a 1-mm digestible chip and a patch that picks up the signal from the chip and can also capture the HR . Another example of recent developments that are particularly significant for COPD and asthma patients is AstraZeneca’s inhaler device called Turbuhaler, which has recently been accompanied by a monitoring device to monitor the actuations of the inhaler .
A related problem is how to tailor medications to individual needs. According to the GOLD: “Each pharmacological treatment regime needs to be adapted to the patient (i.e., individualized), guided by the severity of the symptoms, the risk of exacerbations, comorbidities, drug availability, and the patient’s response” . This can be at least partially achieved by using remote monitoring, for the purpose of titrating/adjusting the treatment, or providing patients with personalized advice. For example, in stable COPD, an increase in FEV1 following a therapeutic trial of corticosteroids for several days is often taken as an indication of regular use for these drugs .
Detection and classification of exacerbations
Signs are objective, whereas symptoms are subjective, evidence of a health problem.
The symptoms of severe COPD exacerbations that require hospitalization are 
Change in cough frequency.
Change in sputum production and appearance.
Increase in dyspnea at rest.
The signs of severe COPD exacerbations that require hospitalization are 
Inability to speak one full sentence.
Temperature > 38.5 °C.
Respiratory rate > 25/min.
HR > 110/min.
PaO2 < 8 kPa .
Use of accessory muscles.
Loss of alertness.
PEF < 100 l/min.
All of these parameters are measurable at home. Parameter 12 is significant on its own, whereas parameters 3, 7, 8, 10, and 11 are significant as a group . It is important to note that the exact thresholds in the previous list are not universally accepted; studies in remote monitoring employ diverse exacerbation criteria , which might be one of the reasons for the inadequate performance of decision support algorithms (discussed in the next section).
The criteria for severe COPD exacerbations based on parameters that are normally measured in hospitals but can now also be measured in the home environment with portable spirometers, transcutaneous measurements, and portable ECG devices are  FEV1 < 1 l, PaO2 < 8 kPa (60 mmHg), SaO2 < 90%, PaCO2 ≥ 6.0 kPa (45 mmHg), and ECG abnormalities. Additional measurements for severe acute exacerbation, which are currently not measurable at home, are chest radiograph, white blood cell count ≥ 12,000, sputum stain/culture, biochemistry (electrolytes, urea, glucose, etc.). The life-threatening events are respiratory or cardiac arrest, confusion or coma, PaO2 < 6.7 kPa (50 mmHg), PaCO2 ≥ 9.3 kPa (70 mmHg), pH < 7.3.
COPD exacerbations can often be prevented . It is therefore desirable to predict or at least early detect signs and symptoms of exacerbations. This can be done automatically by using decision support systems featuring the classification of patient states.
Decision support (exacerbation prediction and detection algorithms)
In most of the existing remote-monitoring systems, the information obtained is analyzed by health caregivers. Only some of them provide automatic decision support systems , stand-alone or in combination with human analysis. Figure 6 presents the COPD-related decision support systems data obtained from two existing reviews [20, 21], and reports featuring decision support that came after the reviews [84, 103, 134,135,136,137]. The most often used inputs are the self-reporting of symptoms on a PCD, followed by pulse oximetry and spirometry (Fig. 6a). ECG has been used in only two publications: as a source of the features for exacerbation prediction , and to detect “clinical alert” (further details not provided) , whereas the PtcCO2 was not used at all.
Panel b shows that only about one quarter of the research featuring decision support provides automatic data acquisition. This is related to the frequency of data acquisition and analysis, which was almost exclusively on a daily basis, except in  where it was 3 h, in  where the interval could be varied depending on each patient’s needs, as well as in , which is the only report featuring continuous data processing. Patients’ compliance is affected by the frequency and method of data acquisition, as discussed in Sect. 4.3, but more importantly, it is not possible to detect immediate life-threatening events without continuous monitoring.
Clinical decisions are traditionally based on a set of predefined universal rules (see previous section). It is for that reason that the approach most often utilized for detecting and predicting exacerbation was by defining universal (population-based) thresholds on obtained symptoms and physiological parameters (panel c). These thresholds were sometimes adjusted to individual patients’ needs, but the best results were obtained by using more advanced classification algorithms: linear discriminant classification , a Bayesian network , a probabilistic neural network classifier , multilevel logistic regression , classification and regression trees , k-means clustering , a state machine combined with logistic regression , and a hybrid classifier combining a support vector machine, random forest, and a rule-based system . The purpose is to classify the patient’s status as being in exacerbation (detection) or as transitional towards exacerbation, i.e., the prodromal period (prediction).
The reported accuracy in detecting exacerbations ranged from 40 to 94%, the sensitivity from 6 to 80%, and the specificity from 61 to 95%. The best accuracy in the early detection of exacerbations is reported for the hybrid classifier with 10 measured parameters and a total of 25 features used as the inputs . Nevertheless, by using only self-reported symptoms as inputs, and k-means clustering, it is possible to obtain a sensitivity of 75% and 90% specificity for early exacerbation detection . Furthermore, a pulse oximeter alone in combination with a classification algorithm can provide a high predictive accuracy .
Only two studies [142, 145] reported using patients’ electronic health records, but these were developed only for the purpose of the study and did not seem to be integrated into the patients’ standard care.
Besides completely outperforming the threshold approach and providing encouraging results, advanced classification algorithms ensure that the classification is adapted to each patient, provide ranking of features based on their predictive power , as well as the extraction of new rules . For a concise description of the predictive analytics methods used in healthcare, the reader can refer to .
In addition to detecting exacerbations, it is beneficial to assess their severity for the purpose of deciding between home or hospital treatment.
Assessment of severity and deciding between home or hospital treatment
The severity of an exacerbation is assessed crudely by tachypnoea, tachycardia, the use of accessory respiratory muscles, cyanosis, and evidence of respiratory muscle dysfunction or fatigue (e.g., uncoordinated ribcage motion or paradoxical movement of the abdominal wall during inspiration) . If the severity of an exacerbation is in doubt, it should always be assessed in hospital. Referral to a hospital’s emergency department is mandatory in the case of respiratory failure indicated by the increased use of accessory respiratory muscles, paradoxical movement of the abdominal wall during inspiration, and significant deterioration in mental status.
Mild exacerbations of COPD are generally believed to represent an increase in symptoms, especially dyspnea, not necessarily accompanied by increased cough and sputum production, which might be more tenacious than usual. These parameters can be obtained through a self-assessment (see Sect. 4.5). Severe exacerbation, on the other hand, is associated with acute respiratory failure, especially in patients with an impaired lung function, sometimes accompanied by hypercapnia , which can be obtained from a PaCO2 measurement. Severe exacerbations cannot be treated at home, so if detected, patients should be transported to a hospital.
There are no definitive clinical guidelines about whether a patient should be cared for at home or in a hospital, and physicians are often uncertain when making this decision. The most important factors are the severity of the exacerbation, acute respiratory failure, the onset of new physical signs (e.g., cyanosis, peripheral edema), and the failure of an exacerbation to respond to initial medical management . Other factors that can be taken into account are cause of the exacerbation (for example, severe pneumonia), a coexisting disorder that requires admission, degree of disability, social factors like the degree of support in the community (e.g., whether the patient lives alone), patient’s history, and mental state [7, 75].
Detection of provoking and predictive factors
It is not clear which factors determine the development and severity of an exacerbation :
It is commonly thought that viral and bacterial infections of the tracheobronchial tree are the major causes of exacerbations in the later stages of disease . The role of bacterial and viral infections in COPD exacerbations is still considered as controversial by some authors .
Air pollution .
There is some evidence that ozone concentration might be slightly associated with additional hospital admissions .
Poor nutrition, i.e., malnutrition, in combination with respiratory muscle fatigue can aggravate the exacerbation.
Drugs (especially tranquilizers).
Stopping regular medication such as diuretics and/or bronchodilators on the patient’s own initiative can increase symptoms, which means that the monitoring of medication compliance is important (Sect. 4.6).
Inappropriate oxygen administration can aggravate an exacerbation because of a reduction in the hypoxic respiratory drive.
Conditions that can mimic or aggravate symptoms are pneumonia, pulmonary hypertension, heart failure or arrhythmias, pulmonary embolism, and pneumothorax.
The cause of about one third of severe exacerbations of COPD cannot be identified .
A study involving 64 patients with moderate-to-severe COPD showed evidence that chronic hypercapnic respiratory insufficiency (high PaCO2) and pulmonary hypertension are predictive factors for hospitalization caused by COPD exacerbation . Long-term oxygen therapy and perhaps even long-term noninvasive mechanical ventilation at home are possibly factors that reduce the risk of severe exacerbations, since there is evidence that they reduce hospital admission in COPD with chronic hypercapnia .
A study using the SGRQ showed that factors for predicting frequent exacerbations were daily cough, daily wheeze (clinical sign of exhaling difficulties caused by a narrowed tracheobronchial tree), and daily cough and sputum together, and frequent exacerbations in the previous year . Another study showed that SGRQ scores and poor quality of life are associated with re-admission for COPD .
Even though there are reports that educational programs for COPD patients can significantly reduce the utilization of healthcare services and improve health status , they have not been as actively promoted as much as programs for asthma patients . Single-topic programs are available (e.g., smoking cessation, long-term oxygen therapy, rehabilitation), but there are insufficient integrated educational materials incorporating all the aspects of disease management .
Educational programs should improve people’s knowledge about the disease process and its treatment and should also motivate patients to change behavior and lifestyle, with the goal of improving their quality of life .