A Markov decision analytical model was built with Microsoft Excel 2016 software to estimate the long-term cost-effectiveness of IoT-based (intervention) versus standard NPPV (control) in patients with COPD and hypercapnic chronic respiratory failure.

This study was based on a multicenter, prospective, randomized controlled trial (RCT; ChiCTR1800019536), which evaluated the effectiveness and safety of IoT-based management of home NPPV for patients with COPD and hypercapnic chronic respiratory failure in China. The trial was conducted over 12 months and involved 148 patients with a mean baseline age of 70.18 years and GOLD grades III-IV [29]. Patients were randomly allocated in a 1:1 ratio to the 2 study groups based on a computer-generated random number sequence implemented by an independent data center [28]. Both groups initiated oxygen therapy and NPPV treatment in the hospital, with an identical ventilator parameter titration protocol during hospitalization, and were discharged after meeting the treatment targets.

In the standard NPPV (control group; n=75), patients were advised to use home oxygen therapy for at least 12 hours per day and home NPPV intermittently for at least 8 hours per day, with continuous NPPV use during sleep, to maintain an oxygen saturation (SpO₂) above the target of 88%.

In the IoT-based NPPV (intervention group; n=73), in addition to the home oxygen therapy for at least 12 hours per day and home NPPV for at least 8 hours per day, with continuous NPPV use during sleep, patients received additional IoT-based management for the full 12 months. The intervention group remotely collected ventilator data via a fourth-generation mobile communication technology (4G) system attached to the back of the ventilator and transmitted the data to the secured platform. The IoT cloud platform (Curative Medical Technology Inc) consists of a data collector, parse server, automatic processing server, data storage center, application interface server, and web server. The platform monitors clinical parameters such as use time, NPPV pressure, mask leaks, breaths per minute, and tidal volume information in real time, and automatically generates daily reports for medical staff. The platform delivers intervention suggestions when adverse reactions, leaks, or poor treatment efficacy occur. The IoT cloud platform triggers automatic alarms when the last 1-week average use is <5 h/night, mask leak is >60 L/min on 3 consecutive nights, or breaths are >25/min on 2 consecutive nights. Health care professionals respond promptly with telephone outreach; if necessary, emergency home visits are arranged, or patients are referred for in-hospital evaluation. Physicians adjust treatment plans based on platform data and patient feedback, with postdischarge management coordinated by family doctors, nurses, and health care providers (Air Liquide Healthcare). Health care professionals conduct in-home follow-ups at 1, 4, and 8 months after treatment, along with monthly telephone follow-ups, to provide health education, monitor the patient’s condition, and support long-term treatment adherence.

During the entire study duration, patients in both groups also received standard COPD treatment, including smoking cessation interventions and medication treatment, in accordance with international treatment guidelines.

Patients who were lost to follow-up, were noncompliant, or refused telemedicine, telephone contacts, or home visits were excluded from this analysis. A total of 125 patients (control group, n=67; intervention group, n=58) were included. Full trial details have been published elsewhere [28]. The baseline characteristics of the study population are shown in Table S1 in Multimedia Appendix 1. The groups appeared well balanced with no major differences in baseline characteristics.

The trial was approved by the medical ethics committee of Zhongshan Hospital, Fudan University (Shanghai, China; B2017-176R) and by local research and development committees at participating centers. In this trial, all patients provided written informed consent, and their data were anonymized. After enrollment, patients were provided with 1 year of free device use and transportation subsidies for 4 follow-up visits. Patients or the public were not involved in our research’s design, conduct, reporting, or dissemination plans.

The model structure was adapted from Zhou et al [30] based on real-world COPD progression, consisting of 4 health states: stable period, nonserious exacerbation period, serious exacerbation period, and death (Figure 1). The stable period health state represents a period of stable health characterized by the absence of exacerbations. Exacerbation-related health states, namely nonserious exacerbation period or serious exacerbation period, were defined by the severity of an exacerbation and whether additional medical resources were needed. Nonserious exacerbations were events that led to patients needing medications adjusted (increasing the type of medication or dosage or changing medication), whereas serious exacerbations were defined as events that resulted in hospitalization. Patients could transition to the death health state from any other health state. Patients remained in one health state in each cycle and then transferred to another health state in the next cycle. The model adopted a lifetime horizon corresponding to 8 years, calculated based on the difference between China’s national average life expectancy in 2019 (77.3 years) and the average age of clinical trial patients included in this analysis (70.04 years). Each year represented a single model cycle, during which the 58 patients in the IoT-based management arm and the 67 patients in the standard management arm could progress throughout the model. The study adheres to Consolidated Health Economic Evaluation Reporting Standards (CHEERS) guidelines.

The RCT followed up with patients at 3, 6, and 12 months, so each patient had 3 periods of follow-up data, including COPD Assessment Test (CAT) scores, medication types, costs, and the number and costs of acute exacerbation hospitalizations. Due to the lack of patients’ baseline medication types, we could not identify whether patients had a medication adjustment in the first 3 months. Therefore, we could not identify whether patients were in the stable or nonserious exacerbation period in the first 3 months; we could only identify whether they were in the serious exacerbation period. The calculation of parameters in this study did not include data from this follow-up for patients whose health state could not be determined.

Only direct costs collected as part of the trial, analyzed from the perspective of the Chinese health care payer, were included in this analysis. Cost components were grouped as either treatment costs, IoT costs, or hospitalization costs and were assumed to account for all aspects of care delivered during each health state (excluding death). Annual costs per patient differed for the IoT-based and standard management groups, as shown in Table 1. All costs are reported in Chinese Yuan (¥). Currency conversion to US dollars was based on the average exchange rate in 2019 (US $1=¥6.9).

Treatment costs included device (oxygen concentrator and NPPV device) setup, maintenance, and support costs, considered one-time costs at the beginning of the treatment period based on standardized costs for the previous 12 months. Rental of ventilators and oxygen supply were also included, estimated based on the purchase price using the annuity method [31] over an assumed 3-year lifespan of the devices. Additionally, medication and follow-up examination costs in the stable and nonserious exacerbation period health states were identified based on patients’ self-reported case report forms and verified through electronic health records, where possible.

The next cost category, IoT costs, covered all costs associated with implementing IoT-based management and included telemedicine equipment maintenance, telephone contact, and health care professional home visits. The costs of telephone consultations and home visits from research and clinical teams were calculated based on hourly rates for different levels of health professionals.

Finally, hospitalization costs, covering all direct costs resulting from exacerbations, including medical and transportation costs, board and lodging, and nutrition and nursing, were collected again based on patient self-reported case report forms.

Quality-adjusted life years (QALYs) were used as effect indicators for this model. Utility values were derived from CAT scores using the following mapping equation (equation 1) [32]. This mapping was developed based on data from patients with COPD in Taiwan, using multiple linear regression with the ordinary least squares method, and was internally validated within the derivation sample, demonstrating good predictive performance.

In equation 1, Utility represents EQ-5D-3L health utility value, where 1 represents full health and 0 represents death, CAT represents CAT scores, and age represents patient age.

Utility values were calculated for each health state, and values used in this analysis are shown in Table 2, where a utility value of 1 represents perfect health, and a value of zero represents death.

Transition probabilities were primarily derived from the clinical trial and supplemented by official Chinese databases. The annual natural mortality rate of the population was obtained from the Chinese and Employment Statistics Yearbook 2021 “National Death Population States by Age and Sex in 2020 (November 1, 2019 to October 31, 2020)” [33] and converted into the natural mortality rate of participants aged 40 to 80 years of whom 84% were male in the clinical trial. This was used to calculate the transition probabilities of patients from the stable period and nonserious exacerbation period health states to death. Other transition probabilities were calculated based on the patient population in the clinical trial. The conversion from incidence to probability was obtained by using the formula (equation 2) [34]. All model transition probabilities are shown in Table 3.

In Equation 2, tp_t represents the probability of event occurrence during t, r means the incidence rate during t, and t represents the study time.

Details of the cost-effectiveness analysis can be found in the Multimedia Appendix 1.

In line with the Chinese Guidelines for Pharmacoeconomic Evaluation (2020) [35], future costs and QALYs were discounted at 5% each year. Model outputs included total per-patient costs and QALYs and an incremental cost-effectiveness ratio (ICER), calculated to express the incremental difference in costs and QALYs between IoT-based management and standard management groups. The ICER was judged against a willingness-to-pay (WTP) threshold equal to per capita gross domestic product (GDP). The 2019 per capita GDP was ¥70,100 in China, according to the official data from the National Bureau of Statistics. IoT-based management was considered cost-effective when the ICER value was lower than ¥70,100 per QALY.

In the base-case analysis, total per-patient costs were ¥94,632.50 for the IoT-based management group and ¥91,025.24 for the standard management group; total QALYs were 3.49 and 3.25, respectively, across the model time horizon. Compared with the standard management group, IoT-based management increased costs by ¥3607.26 but also increased QALYs by 0.24 per patient across the time horizon, resulting in an ICER of ¥15,030.25 per QALY. As the ICER was substantially less than the WTP threshold of ¥70,100 per QALY, IoT-based management was cost-effective compared with standard management. These results are presented in Table 4.

Results of the OWSA, displayed in Figure 2, showed that the parameter that drove the greatest change in ICER results was the hospitalization cost of a serious exacerbation for the standard management group. The second most influential parameter was the hospitalization cost of a serious exacerbation for the IoT-based management group. All other parameter changes had comparatively less influence on model results.

Results of the PSA, presented in Figure 3, showed the uncertainty around the model’s cost and QALY estimates. A total of 93.6% of iterations fell below the WTP threshold (indicated by the dotted line), indicating that IoT-based management had a 93.6% probability of being cost-effective relative to standard management at the given WTP threshold. This relationship between the probability of being cost-effective as the WTP threshold changes is shown in Figure 4. Additionally, in 80.2% of PSA iterations, IoT-based NPPV was more effective (accumulating more QALYs) and cheaper than standard NPPV.