Highlights
- •Multi-year retrospective cohort study of patients with COPD-CRF receiving Non-invasive Ventilation at Home (NIVH) using 100% Medicare claims.
- •Econometric IPTW matching methodology – Cox proportional hazard models for time-to-event outcomes and generalized least square model for Medicare expenditure outcomes.
- •Findings are that hypercapnic patients benefited in reduced mortality risk from NIVH applied within the first week after index; hypoxic patients did not reduce their risk of mortality.
- •Medicare expenditures for hypercapnic patients were lower in the 12 months after treatment than expenditures for either hypoxic or unspecified phenotypes.
Abstract
Background
While non-invasive ventilation at home (NIVH) is gaining wider acceptance as a treatment option for chronic obstructive pulmonary disease with chronic respiratory failure (COPD-CRF), uncertainty remains about the optimal time to begin NIVH, whether a specific phenotype of COPD-CRF predicts improved outcomes, and how NIVH affects healthcare costs.
Materials and methods
Using 100% research identifiable fee-for-service Medicare claims from 2016 through 2020, we designed an observational, retrospective, cohort study to determine how NIVH use in COPD-CRF patients stratified by CRF phenotype and by timing of initiation affected mortality, healthcare utilization, and total healthcare costs compared to a matched control group.
Results
In hypercapnic COPD-CRF patients starting NIVH within the first week following diagnosis, risk of death was reduced by 43% (HR, 0.57; 95% CI 0.51–0.63, p < .0001), those starting 8–15 days following diagnosis had mortality reduction of 31% (HR, 0.69; 95% CI 0.62–0.77, p < .0001), and those starting 16–30 days following diagnosis showed mortality reduction of 16% (HR 0.84, CI 0.073-0.096, p < .01) compared to controls. Medicare spending was also associated with timing of NIVH initiation in hypercapnic COPD-CRF. Those beginning treatment 0–7 days and 0–15 days following diagnosis had a $5484 and a $3412 reduction in Medicare expenditures respectively the next year. NIVH was not associated with improved clinical outcomes or decreased Medicare spending in COPD-CRF patients who were not hypercapnic.
Conclusion
In this study, early initiation of NIVH for hypercapnic COPD-CRF patients was associated with reductions in the risk of death and in total Medicare spending.
Keywords
Abbreviations:
chronic obstructive pulmonry disease (COPD), chronic respiratory failure (CRF), non-invasive ventilation at home (NIVH)1. Introduction
Chronic obstructive pulmonary disease (COPD) affects more than 20 million Americans and is the fourth leading cause of death in the United States (U.S.), responsible for an estimated 120,000 deaths per year [
[1]
]. Caring for COPD patients in the U.S. is resource intensive, accounting for 700,000 hospitalizations, 1.5 million emergency department (ED) visits, and 50 billion dollars in annual cost to society [[2]
].COPD has no cure, is often progressive, and may lead to the development of chronic respiratory failure (CRF), which is defined by a combination of symptoms, frequent exacerbations, chronic hypoxia, and/or chronic hypercapnia [
[3]
,[4]
]. Patients with COPD-CRF have a poor quality of life, high morbidity and mortality, and consume a disproportionate share of healthcare resources. Therapies for COPD-CRF are limited and primarily supportive with goals of improving symptoms, decreasing mortality, and reducing healthcare utilization [[5]
,[6]
].As evidenced by a recent clinical practice guideline, non-invasive ventilation at home (NIVH) is gaining wider acceptance as a treatment option for COPD-CRF [
[7]
]. Despite emerging evidence of benefit and an increase in NIVH utilization, it remains an infrequently used treatment with less than 5% of COPD-CRF patients receiving this therapy [[8]
].A meta-analysis published in 2020 concluded NIVH use in COPD-CRF reduced hospitalizations but that it did not reduce mortality [
[9]
]. In comparison, two retrospective cohort studies published in 2021 demonstrated NIVH use in COPD-CRF patients was associated with significant reductions in mortality, hospitalizations, and emergency department visits [[8]
,[10]
]. These latter findings are in line with results from small, randomized trials of NIVH in COPD-CRF performed in Europe [[11]
,[12]
].To date, no U.S. based studies have been published exploring the effects of NIVH on total healthcare costs for COPD-CRF patients. Additionally, uncertainty remains regarding the optimal time to begin NIVH during the clinical course of COPD-CRF, as well as whether the presence of a specific phenotype of CRF better predicts improved clinical outcomes with NIVH therapy [
[7]
,[13]
].- Gay P.C.
- Owens R.L.
On behalf of the ONMAP technical expert panel, executive summary: optimal noninvasive ventilation Medicare access promotion (ONMAP).
Chest. 2021; https://doi.org/10.1016/j.chest.2021.05.074
To help answer these questions, we designed an observational, retrospective cohort study using research identifiable Medicare claims from 2016 through 2020. Our primary objectives were defining how NIVH use stratified by COPD-CRF phenotype and by timing of initiation affected mortality, healthcare utilization, and total healthcare costs.
2. Research questions
We sought to answer the following three research questions:
- •What is the optimal time to begin NIVH treatment during the clinical course of COPD-CRF?
- •Is a specific phenotype of COPD-CRF associated with improved outcomes with NIVH use?
- •What is the impact of NIVH treatment on Medicare expenditures?
3. Materials and methods
3.1 Data source
The data used in this study were from the Medicare 100% research identifiable files (RIF) fee-for-service (FFS) and Medicare Beneficiary Summary File (MBSF) administrative claims under Data Use Agreement (DUA) 54757. Data from January 2016 through June 2020 were extracted. The RIF claims files include inpatient and outpatient hospital, skilled nursing facility, hospice, home health agency, physician, and durable medical equipment claims. The MBSF includes beneficiaries’ entitlement and eligibility information, enrollment, and socio-demographic data (date of birth, age, gender, race, date of death if applicable, and state of residence).
3.1.1 Study cohort
The study cohort comprised Medicare beneficiaries who were concurrently diagnosed with COPD and CRF during the period of January 2016 through December 2019. Beneficiaries were identified using International Classification of Diseases, Tenth Revision, Clinical Modification [ICD-10-CM] codes J96.10, J96.20, J96.11, J96.21, J96.12, J96.22 for CRF, and ICD-10-CM codes J40, J41.1, J41.8, J42, J43.0, J43.1, J43.2, J43.8, J43.9, J44.0, J44.1, J44.9 for COPD. The index date was the date of the claim when both COPD and CRF first appeared.
We excluded patients with dementia (ICD-10-CM codes F00.x-F03.x, F05.1, G30.x, G31.1) since such patients often have difficulty tolerating NIVH [
[14]
]. We also excluded patients with obstructive sleep apnea (OSA) (ICD-10-CM G47.33) since NIVH is an effective treatment for this disorder and we wanted to ensure that any observed benefits were due to treatment of COPD-CRF and not OSA [[15]
].To allow the full scope of healthcare services utilization, costs, and comorbidities to be captured in our analysis, beneficiaries were excluded if they were not enrolled in Medicare FFS Parts A and B during the 12-months before and after the index date, or if they were enrolled in a Medicare Part C plan at any time during this same 2-year period. Beneficiaries who had no Medicare expenditures during the 12 months before or after the index date were also excluded from our analysis. Fig. 1A in the Appendix presents the different steps and exclusions of the sample selection.
The sample was divided into a treatment group who received NIVH (Healthcare Common Procedure Coding System [HCPCS] E0466) within two months of the combined COPD-CRF diagnosis, and a control group who did not receive NIVH at any point in the follow-up period. The treatment and control groups were then assigned to one of three COPD-CRF phenotypes based on ICD-10-CM coding. The 3 phenotypes were COPD-CRF unspecified (J96.10, J96.20), COPD-CRF with hypoxia (J96.11, J96.21), and COPD-CRF with hypercapnia (J96.12, J96.22). These phenotypes were mutually exclusive so that each subject was only assigned to one group. Each treatment group member was then assigned to one of four treatment initiation windows defined as the time elapsed between the diagnosis of COPD-CRF and the start of NIVH. These time windows were 0–7 days, 8–15 days, 16–30 days, and 31–60 days.
3.1.2 Outcome variables
The primary study outcome was all-cause mortality (time-to-death). The secondary outcomes were time to first hospital admission, and time to first emergency department (ED) visit after the index date. We also estimated total Medicare expenditures during the 12 months following the index date. Patients were followed until death or for one year if they remained alive.
3.1.3 Statistical analysis
We conducted regression analyses using Cox proportional hazard models for time-to-event outcomes (i.e., time-to-death, time to first hospital admission, and time to first ED visit after the index date). We estimated a generalized linear model for the impact on Medicare expenditures in the next year for the treatment group (NIVH use) versus the control group (non-NIVIH use).
Throughout the modeling, different techniques were used to appropriately answer each research question, while also addressing different forms of bias such as: 1) confounding factors, 2) selection bias, and 3) immortal time bias.
Addressing Potential Biases: Potential confounding factors were addressed by adjusting the different models using numerous socio-demographic and clinical factors. The sociodemographic characteristics included race/ethnicity, gender, age, Medicare/Medicaid dual eligibility, and region of the country where subjects lived. The clinical and health-related factors were the conditions comprising the Charlson Comorbidity Index (to control for the presence of chronic diseases) and total Medicare expenditures in the 12 months before the index date.
Because of the observational nature of the study, we used an inverse probability of treatment weighting (IPTW) propensity score model to address selection bias. Propensity score methods are statistical techniques widely used in observational studies to mitigate self-selection bias [
[16]
]. This approach synthetically creates randomness in the incidence of treatment to address the problem of self-selection between treatment and control individuals without losing any observations from the dataset. Previously published literature suggested that applying these techniques to observational studies sufficiently addresses selection bias such that the results approximate those of randomized clinical trials [[17]
].Although there are several methods of matching patient cohorts, all rely on the propensity score to estimate the probability of treatment assignment conditional on observed baseline covariates using logistic regression. IPTW requires the construction of weights using propensity scores, and then estimating a weighted regression using the constructed weights. Along with the Cox proportional hazards model, IPTW was applied to estimate the three time-to-event outcomes. IPTW was also applied to a generalized linear regression model with a gamma distribution and a log link function to estimate the Medicare expenditures outcome.
For all time-to-event models, we addressed immortal time bias by modeling each beneficiary in the treatment group as a control group member for the time elapsed between the diagnosis date and the treatment initiation date, and as a treatment group member thereafter. We also considered alternative specifications of the Cox models that accounted for the time dependency of the treatment. We used a generalized linear model with gamma distribution and a log link function to estimate total Medicare expenditures. Generally, we use GLM with Log-link function and Gamma Distribution because of non-negative and positively skewed distribution of spending data.
Determining the optimal time to begin NIVH treatment: To determine whether early initiation of NIVH affected survival, we estimated a Cox proportional hazard model with four service receipt categories (0–7 days, 8–15 days, 16–30 days, and 31–60 days). Each service receipt category represented the elapsed time between the index date (i.e., first instance of a combined COPD-CRF diagnosis) and the NIVH treatment initiation date. Based on these results, we constructed additional models with binary treatment indicators that covered three treatment initiation windows (0–7 days, 0–15 days, and 0–30 days).
We also investigated whether the different time-to-treatment windows affected all cause ED visits, all cause hospitalizations, and Medicare expenditures by running the appropriate models for the secondary outcome variables using the different treatment windows. In the case of binary treatment, given that the estimating effect represents an average over two or three time periods, we expected that if late treatment was less effective, the estimated effect would decrease with longer treatment initiation windows.
That is, if the benefit in the second week was less than the benefit in the first week, and if both weeks were combined (expanded treatment initiation window), the average of the two weeks combined would be less than the average in the first week. Ideally, we would use categorized treatment windows for all models; however, we opted for cumulative windows that allowed us to use binary, yes or no, treatment indicators to implement the IPTW to address selection bias, while also accounting for immortal time bias.
Determining whether a specific phenotype of COPD-CRF was associated with improved outcomes- Subsample analyses: In addition to estimating our different models using the full samples, we conducted subsample analyses to assess whether different COPD-CRF phenotypes responded differently to NIVH. Specifically, we divided the patients into 3 groups using the ICD-10-CM coding information contained in the RIF and ran the time -to-event models on the phenotype subsamples described above. We then ran the model for total Medicare expenditures on these same COPD-CRF phenotype subsamples.
For each of these models, we assessed how well the reweighting improved the balancing of the baseline characteristics by reporting the absolute values of the standardized mean differences. A common rule in the literature is that the standardized mean differences should be within the interval of [0, 1] for a well-balanced baseline [
[18]
]. We examined these values using “love plots” for the full sample and for the full sample at treatment window 0–30 days.4. Results
The statistical results are organized as follows. We first show the distribution of the study population, followed by the summary statistics of the covariates used in this analysis. The different characteristics (demographics, geographic location and patient health) of treatment and control groups are well balanced after the IPTW propensity score match; the groups are thus similar in terms of all of the characteristics. Finally, we present the regression results.
4.1 Baseline characteristics of the study sample
Fig. 1 (in the Appendix) reports the selection process of the study sample, showing all exclusions.
Table 1 shows the distribution of patients according to NIVH use, COPD-CRF phenotype, and by time elapsed between the index date and treatment initiation. The final sample consisted of 499,717 patients with COPD-CRF of which 6707 (1.3%) were prescribed NIVH (treatment group) and 493,010 (98.7%) were not (control group). The sample was further divided into 3 subsamples based on the ICD-10-CM coding for COPD-CRF phenotype.
Table 1Distribution of study population.
| Total Study Sample | Hypercapnic Subsample | Hypoxic Subsample | Unspecified | |||||
|---|---|---|---|---|---|---|---|---|
| Count | Percentage | Count | Percentage | Count | Percentage | Count | Percentage | |
| Never Used NIVH | 493,010 | 98.70% | 76,371 | 95.50% | 314,623 | 99.40% | 102,016 | 98.90% |
| NIVH within 60 Days | 6707 | 1.30% | 3629 | 4.50% | 1941 | 0.60% | 1137 | 1.10% |
| Time To Treatment | ||||||||
| 00–07 Days | 1962 | 0.40% | 1143 | 1.40% | 459 | 0.14% | 360 | 0.35% |
| 08–15 Days | 1648 | 0.30% | 955 | 1.20% | 401 | 0.13% | 292 | 0.28% |
| 16–30 Days | 1364 | 0.30% | 666 | 0.80% | 459 | 0.14% | 239 | 0.23% |
| 31–60 Days | 1733 | 0.30% | 865 | 1.10% | 622 | 0.19% | 246 | 0.24% |
| Total | 499,717 | 100% | 80,000 | 100.00% | 316,564 | 100.00% | 103,153 | 100.00% |
The first subsample was the 80,000 patients with hypercapnic COPD-CRF (J96.12 and J96.22), of which 3629 (4.5%) received NIVH. These patients comprised 16% of the total sample, 15% of the control group and 54% of the NIVH treatment group. The second subsample was the 316,564 patients who had hypoxic COPD-CRF (J96.11 and J96.21) of which 1941 (0.6%) received NIVH. These patients comprised 63% of the total sample, 64% of the control group, and 29% of the NIVH treatment group.
The third subsample was the 103,153 patients with COPD-CRF unspecified (J96.10 and J96.20) of which 1137 (1.1%) received NIVH. These patients comprised 21% of the total sample, 21% of the control group and 17% of the NIVH treatment group. We analyzed how each of these subsamples responded to NIVH treatment as compared to untreated control groups with the same COPD-CRF phenotype.
Table 2 contains the sociodemographic and clinical characteristics of the overall study sample. The unweighted summary statistics show significant differences between the NIVH treatment group and the control group. For example, the NIVH treatment group was younger, had lower prior 12-month healthcare spending, and were more likely to have COPD-CRF with hypercapnia than those in the control group. The Charlson Comorbidity conditions all occurred more commonly in the control group than in the treatment group, except for a chronic pulmonary disease diagnosis occurring after a CRF diagnosis, which occurred in 91% of the treatment group and 85% of the control group.
Table 2Characteristics of the study population.
| NIVH Use | All | P-values | ||
|---|---|---|---|---|
| Characteristics | No | Yes | ||
| Patient count | 493,010 | 6707 | 499,717 | |
| Age | 75.83 ± 10.53 | 72.13 ± 9.74 | 75.78 ± 10.53 | 0.0001 |
| Female | 56.50% | 57.20% | 56.50% | 0.2640 |
| Race-White | 88.70% | 89.40% | 88.70% | 0.0590 |
| Race-Black | 7.10% | 6.50% | 7.10% | 0.0290 |
| Race-Asian | 1.00% | 1.10% | 1.00% | 0.5060 |
| Race-Hispanic | 1.10% | 0.90% | 1.10% | 0.0570 |
| Race- American Indian/Pacific Islander | 0.70% | 0.70% | 0.70% | 0.8160 |
| Race -Other | 1.40% | 1.40% | 1.40% | 0.6940 |
| Dual Medicare/Medicaid | 25.90% | 27.60% | 25.90% | 0.0020 |
| ESRD | 4.90% | 1.70% | 4.90% | 0.0001 |
| Inpatient Index [1] | 37.20% | 34.60% | 37.10% | 0.0001 |
| Region 1: CT, ME, MA, NH, RI, VT | 5.00% | 2.90% | 4.90% | 0.0001 |
| Region 2: NY, NJ, PR | 7.00% | 7.30% | 7.20% | 0.7270 |
| Region 3: DE, MD, DC, WV, VA, PA | 11.00% | 9.20% | 10.50% | 0.0001 |
| Region 4: NC, SC TN, FL, GA, AL, KY, MS | 23.00% | 25.00% | 23.40% | 0.0020 |
| Region 5: MI, MN, OH, IL, IN, WI | 18.00% | 14.50% | 17.90% | 0.0001 |
| Region 6: TX, LA, AR, OK, NM | 12.00% | 17.80% | 12.50% | 0.0001 |
| Region 7: MO, KS, IA, NE | 6.00% | 5.80% | 5.80% | 0.8780 |
| Region 8: ND, UT, SD, WY, CO, MT | 4.00% | 4.40% | 4.00% | 0.1130 |
| Region 9: NV, AS, AZ, CA, GU, HI, MIS | 10.00% | 9.70% | 10.30% | 0.1180 |
| Region10: AK, ID, OR, WA | 3.00% | 3.50% | 3.50% | 0.8610 |
| Myocardial infarction | 19.10% | 12.50% | 19.00% | 0.0001 |
| Malignancy, except neoplasm of skin | 24.90% | 17.00% | 24.80% | 0.0001 |
| Cerebrovascular disease | 25.40% | 16.80% | 25.30% | 0.0001 |
| Congestive heart failure | 46.80% | 35.90% | 46.60% | 0.0001 |
| Chronic pulmonary disease | 85.20% | 90.50% | 85.30% | 0.0001 |
| Diabetes | 35.20% | 29.50% | 35.10% | 0.0001 |
| Diabetes with complications | 22.30% | 14.70% | 22.20% | 0.0001 |
| Hemiplegia, paraplegia | 4.00% | 2.50% | 3.90% | 0.0001 |
| Metastatic solid tumor | 8.30% | 4.00% | 8.20% | 0.0001 |
| Mild liver disease | 10.50% | 6.70% | 10.50% | 0.0001 |
| Moderate/severe liver disease | 1.70% | 0.70% | 1.70% | 0.0001 |
| Peptic ulcer disease | 4.10% | 2.40% | 4.00% | 0.0001 |
| Peripheral vascular disease | 41.60% | 30.80% | 41.50% | 0.0001 |
| Renal disease | 33.60% | 19.40% | 33.50% | 0.0001 |
| Rheumatologic disease | 7.60% | 5.90% | 7.60% | 0.0001 |
| AIDS/HIV | 0.40% | 0.20% | 0.40% | 0.0001 |
| Chronic Hypercapnia with or without chronic hypoxia | 15.50% | 54.10% | 16.00% | 0.0001 |
| Chronic Hypoxia without chronic hypercapnia | 63.80% | 28.90% | 63.30% | 0.0001 |
| Log (Prior 12 months Total Spending) | 9.82 ± 1.44 | 9.34 ± 1.37 | 9.82 ± 1.44 | 0.0001 |
(1)Inpatient index is a flag that is coded 1 if the index COPD-CRF diagnosis was made during an inpatient stay.
(2)Chronic conditions comprise the Charlson Comorbidity Index.
An analysis of the characteristics of the study population by CRF phenotype found that hypercapnic patients tended to be more likely black, dually eligible for Medicare and Medicaid, diabetic, and to have received the COPD-CRF diagnosis during an inpatient stay. All differences were highly significant (p < .0001). These unadjusted results can be found in Table 1A in the Appendix.
4.2 Outcome results
4.2.1 Primary end point: time-to-death
Table 3 contains the hazard ratios for mortality associated with the different treatment initiation windows and different COPD-CRF phenotypes corrected for immortal time bias. Risk reduction for death associated with NIVH treatment in the full sample and the hypercapnic subsample significantly decreases as the time between diagnosis and treatment initiation increases. This suggests that the earlier a hypercapnic patient starts NIVH treatment, the greater the reduction in mortality. Of note, no mortality reduction was found in the hypoxic or unspecified COPD-CRF patients treated with NIVH in any treatment initiation window. In fact, hypoxic COPD-CRF patients had a significantly increased mortality risk in the 16–30 days and 31–60 days treatment initiation windows. Given that the hypercapnic subsample, comprising 54% of the full sample, is the only one associated with NIVH benefit, it is this subsample that is responsible for the benefit seen in the full sample analysis.
Table 3Time-to-death models by ordinal treatment window results.
| Full Sample | Hypercapnic Subsample | Hypoxic Subsample | Unspecified | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Treatment Window | HR | Confidence Limits | p- | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | ||||
| values | ||||||||||||||||
| 0-7days | 0.70 | 0.64 | 0.76 | 0.0001 | 0.57 | 0.51 | 0.64 | 0.0001 | 1.01 | 0.84 | 1.35 | 0.6064 | 0.88 | 1.08 | 1.46 | 0.2280 |
| 8–15days | 0.76 | 0.69 | 0.83 | 0.0001 | 0.69 | 0.62 | 0.77 | 0.0001 | 1.02 | 0.90 | 1.28 | 0.4391 | 0.85 | 1.07 | 1.87 | 0.1720 |
| 16–30days | 0.92 | 0.84 | 1.02 | 0.0988 | 0.84 | 0.73 | 0.96 | 0.0110 | 1.23 | 1.01 | 1.34 | 0.0404 | 1.03 | 0.67 | 1.07 | 0.8280 |
| 31–60 days | 1.10 | 1.01 | 1.19 | 0.0288 | 1.01 | 0.89 | 1.13 | 0.9152 | 1.54 | 1.34 | 1.76 | <.0001 | 1.18 | 0.92 | 1.50 | 0.1866 |
Note: Estimates are corrected for immortal time bias.
Table 4 contains results of the regression analysis for the full sample and the different COPD-CRF phenotypes and three treatment windows, and accounts for immortal time bias. Regardless of the specifications, NIVH was associated with improved survival in the full sample and in patients with hypercapnia. The immortal time bias analysis for the three treatment initiation windows and COPD-CRF phenotypes showed that NIVH became less effective in reducing risk of death in both the full sample and in hypercapnic COPD-CRF the longer the delay in treatment initiation (see the above comments regarding the effects of the hypercapnic sample on the full sample). As in Table 3, the hypoxic COPD-CRF group had an increased mortality risk if NIVH was started 0–30 days following diagnosis.
Table 4Death risk difference from cox proportional hazard models.
| Full Sample | Hypercapnic Subsample | Hypoxic Subsample | Unspecified | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Treatment Window | HR | Confidence Limits | p- | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | ||||
| values | ||||||||||||||||
| 0–7 days | 0.69 | 0.62 | 0.76 | 0.0001 | 0.57 | 0.50 | 0.66 | 0.0001 | 1.06 | 0.84 | 1.35 | 0.0606 | 0.97 | 0.73 | 1.28 | 0.8080 |
| 0–15 days | 0.74 | 0.68 | 0.81 | 0.0001 | 0.64 | 0.57 | 0.71 | 0.0001 | 1.07 | 0.90 | 1.28 | 0.7965 | 0.92 | 0.75 | 1.14 | 0.4664 |
| 0–30 days | 0.82 | 0.76 | 0.88 | 0.0001 | 0.70 | 0.64 | 0.77 | 0.0110 | 1.16 | 1.16 | 1.34 | 0.0404 | 0.98 | 0.8 | 1.11 | 0.7069 |
Note: Estimates are corrected for immortal time bias.
4.3 Outcome results
Secondary End Points:
Time-to-First Inpatient Admission and Time-to-First ED Visit after Index Date.
Table 5 presents the results of the Cox proportional hazard analysis for time-to-first hospitalization following the index date. Among patients with hypercapnia, NIVH reduced the risk of hospitalization. This model corrected for immortal time bias, was adjusted and IPTW reweighted, and showed all time to treatment initiation windows had comparable, statistically significant reductions in hospitalizations. Initiating NIVH between 0 and 7, 0–15, or 0–30 days, led to decreases in the risk of hospitalizations of approximately 23%. There was no effect of NIVH on the risk of hospitalizations in patients with hypoxic or unspecified COPD-CRF, regardless of the treatment initiation window.
Table 5Time-to-first inpatient admission after index date.
| Full Sample | Hypercapnic Subsample | Hypoxic Subsample | Unspecified | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Treatment Window | HR | Confidence Limits | p- | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | ||||
| values | ||||||||||||||||
| 0–7 days | 0.92 | 0.84 | 1.01 | 0.0670 | 0.85 | 0.75 | 0.96 | 0.0088 | 1.05 | 0.87 | 1.28 | 0.6013 | 1.08 | 0.87 | 1.34 | 0.5049 |
| 0–15 days | 0.91 | 0.85 | 0.98 | 0.0110 | 0.85 | 0.78 | 0.93 | 0.0007 | 1.09 | 0.94 | 1.25 | 0.2519 | 0.96 | 0.82 | 1.13 | 0.6389 |
| 0–30 days | 0.89 | 0.84 | 0.95 | 0.0002 | 0.85 | 0.78 | 0.92 | 0.0001 | 0.98 | 0.88 | 1.10 | 0.7573 | 0.97 | 0.85 | 1.12 | 0.7102 |
Note: Estimates are corrected for immortal time bias.
Table 6 contains the results for time-to-first ED Visit following the index date. NIVH reduced the risk of ED visits for patients with hypercapnic COPD-CRF who begin treatment between 0 and 30 days, but not in the 0–7 day or 0- 15-day windows. No reduction in ED visits was seen in any treatment initiation window in unspecified or hypoxic COPD-CRF.
Table 6Time-to-first ED visit after index date.
| Full Sample | Hypercapnic Subsample | Hypoxic Subsample | Unspecified | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Treatment Window | HR | Confidence Limits | p- | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | HR | Confidence Limits | P-values | ||||
| values | ||||||||||||||||
| 0–7 days | 0.96 | 0.89 | 1.05 | 0.3777 | 0.92 | 0.83 | 1.03 | 0.1448 | 1.05 | 0.89 | 1.25 | 0.5645 | 1.10 | 0.91 | 1.33 | 0.3116 |
| 0–15 days | 0.96 | 0.90 | 1.02 | 0.1498 | 0.92 | 0.85 | 1.00 | 0.0582 | 1.06 | 0.93 | 1.20 | 0.3771 | 1.00 | 0.87 | 1.16 | 0.9974 |
| 0–30 days | 0.94 | 0.89 | 0.99 | 0.0153 | 0.92 | 0.86 | 0.99 | 0.0209 | 0.99 | 0.90 | 1.10 | 0.8969 | 0.98 | 0.86 | 1.11 | 0.7090 |
Note: Estimates are corrected for immortal time bias.
4.4 Outcome results
4.4.1 Secondary end point: Medicare expenditures
Table 7 presents the Medicare healthcare expenditures for the year following the diagnosis of COPD-CRF. Results are shown for both the NIVH treated group and the control group as a function of time to initiation of NIVH, if it was started, and for the phenotype of CRF.
Table 7Effect of NIVH on medicare expenditures.
| Full Sample | Hypercapnic Subsample | Hypoxic Subsample | Unspecified | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Treatment Window | Treatment | Healthcare Costs for following year | Standard Error | Pr > ChiSq | Healthcare Costs for following year | Standard Error | Pr > ChiSq | Healthcare Costs for following year | Standard Error | Pr > ChiSq | Healthcare Costs for following year | Standard Error | Pr > ChiSq |
| 0–7days | NIVH | $41,143 | 101.49 | 0.0001 | $41,598 | 234.23 | 0.0001 | $39,565 | 6194.20 | 0.0001 | $38,546 | 330.63 | 0.0001 |
| 0–7days | Comparison | $43,209. | 106.88 | 0.0001 | $47,082 | 259.28 | 0.0001 | $38,050 | 6194.18 | 0.0001 | $38,522 | 334.58 | 0.0001 |
| 0–7days | Difference | <$2066> | 100.06 | 0.0001 | -<$5484> | 279.29 | 0.0001 | $1515 | 138.24 | 0.0001 | $24 | 273.41 | 0.9812 |
| 0–15days | NIVH | $42,983 | 102.69 | 0.0001 | $43,944 | 241.87 | 0.0001 | $44,069 | 141.75 | 0.0001 | $36,987 | 291.21 | 0.0001 |
| 0–15days | Comparison | $43,598 | 104.55 | 0.0001 | $47,356 | 255.63 | 0.0001 | $38,291 | 118.95 | 0.0001 | $38,171 | 302.59 | 0.0001 |
| 0–15days | Difference | -<$615> | 101.47 | 0.0001 | <$3412> | 282.12 | 0.0001 | $5778 | 139.68 | 0.0001 | <$1184> | 261.37 | 0.0001 |
| 0–30days | NIVH | $46,357 | 108.53 | 0.0001 | $48,090 | 263.66 | 0.0001 | $46,398 | 144.7 | 0.0001 | $39,589 | 296.92 | 0.0001 |
| 0–30days | Comparison | $43,699 | 102.61 | 0.0001 | $47,737 | 258.12 | 0.0001 | $38,346 | 115.98 | 0.0001 | $38,222 | 283.85 | 0.0001 |
| 0–30days | Difference | $2658 | 106.15 | 0.0001 | $352.18 | 296.35 | 0.0001 | $8052 | 144.79 | 0.0001 | $1366 | 263.79 | 0.0001 |
In the hypercapnic COPD-CRF group, starting NIVH in the 0–7 day or 0-15-day windows resulted in significant reductions in Medicare spending of $5484 (p < .0001) and $3412 (p < .0001) respectively in the year following diagnosis. Beginning treatment in the 0–30-day window resulted in no significant change in spending compared to the untreated patients. These results suggest that, as with mortality reduction, the sooner NIVH is begun following a diagnosis of hypercapnic COPD-CRF, the greater are the healthcare cost savings.
Irrespective of treatment initiation window, NIVH use in hypoxic COPD-CRF patients resulted in a significant (p < .0001) increase in Medicare expenditures. For unspecified patients, if NIVH was initiated in the 8–15 day window, there was a reduction in Medicare expenditures of $1184 (p < .0001), but if NIVH was initiated in either of the other two windows, increases in Medicare expenditures were observed.
5. Discussion
This is the third Medicare claims data study showing significant clinical benefits using NIVH in COPD-CRF patients [
[8]
,[10]
], and the first to demonstrate cost savings associated with NIVH use. Additionally, this is the first study to report outcomes as a function of COPD-CRF phenotype and timing of NIVH initiation.Since one important physiological effect of mechanical ventilation is to lower pCO2, many experts suggest limiting NIVH in COPD-CRF to hypercapnic patients theorizing that only these patients will benefit [
[7]
,11
, 12
, 13
].While this recommendation makes sense, we are unaware of any published data empirically investigating this idea. Additionally, the optimal time to prescribe NIVH for COPD-CRF is controversial with some experts suggesting waiting several weeks after an exacerbation before beginning NIVH while others favor immediate treatment. This uncertainty is underscored by a recent practice guideline for NIVH use in COPD-CRF [- Gay P.C.
- Owens R.L.
On behalf of the ONMAP technical expert panel, executive summary: optimal noninvasive ventilation Medicare access promotion (ONMAP).
Chest. 2021; https://doi.org/10.1016/j.chest.2021.05.074
[7]
] and a technical expert panel on the same topic [[13]
] coming to different conclusions. We therefore designed our study to investigate how the timing of initiation of NIVH affected the outcomes of death, hospitalization risk, ED utilization, and Medicare expenditures in different phenotypes of COPD-CRF.- Gay P.C.
- Owens R.L.
On behalf of the ONMAP technical expert panel, executive summary: optimal noninvasive ventilation Medicare access promotion (ONMAP).
Chest. 2021; https://doi.org/10.1016/j.chest.2021.05.074
5.1 Mortality
In both the phenotype and time to treatment initiation analysis, only the hypercapnic COPD-CRF group had a reduction in mortality associated with NIVH. Specifically, hypercapnic COPD-CRF patients begun on NIVH within 0–7 days, 8–15 days and 16–30 days of diagnosis had significant reductions in the risk of death with HR of 0.57 (CI 0.51–0.64, p < .0001), 0.69 (CI 0.62–0.77, p < .0001), and 0.84 (CI 0.73–0.96, p = .01) respectively, suggesting that the earlier this group started NIVH, the more they benefitted. Conversely, neither the hypercapnic patients started on NIVH between days 31–60 nor the unspecified or hypoxic patients started on NIVH during any treatment window had a reduction in mortality.
5.2 Hospitalizations and ED visits
In hypercapnic COPD-CRF patients, NIVH started at 0–7 days, 0–15 days, and 0–30 days following diagnosis, was associated with a significant, similar reduction in the risk of hospitalization of approximately 23%. As with the mortality reduction, if NIVH was begun more than 30 days following diagnosis, no reduction in hospitalizations was seen in the hypercapnic subsample and no reduction in hospitalizations in the unspecified and hypoxic COPD-CRF subsample occurred regardless of when treatment began.
Furthermore, only hypercapnic COPD-CRF patients begun on NIVH between 0 and 30 days had a reduction in ED visits with a HR of 0.94 (CI 0.89–0.99, p = .020). Those started between 0-7 days and 0–15 days had no significant reduction in ED visits with HR of 0.96 (CI 0.90–1.02) and 0.94 (CI 0.89–1.00) respectively. These results suggest that hypercapnic COPD-CRF patients begun on NIVH shortly after diagnosis may have a reduction in mortality at the expense of higher ED use. However, these ED visits do not result in increased hospitalizations since hospitalizations are significantly reduced in hypercapnic COPD-CRF treated with NIVH within 30 days of diagnosis. One possible explanation for this observation may be that NIVH improves patients’ clinical condition so that an ED visit, instead of a hospitalization, provides a sufficient level of care in the event of an exacerbation.
5.3 Economic outcomes
The economic impact on the Medicare program of NIVH use in COPD-CRF has never been reported. To address this, we controlled for immortal time bias and analyzed total Medicare expenditures of Part A (covering inpatient hospital, skilled nursing facility, hospice, lab tests, and home health care) and Part B (covering health care providers' services, outpatient care, and durable medical equipment) in the NIVH group compared to the control group.
As with the clinical benefits, the economic benefit of reduced total cost of care was greatest in hypercapnic COPD-CRF patients and was most pronounced the earlier treatment was begun. For hypercapnic COPD-CRF patients, Medicare expenditures for the year following NIVH initiation was decreased by $5484 (11.6%) compared to controls if treatment was begun within 7 days of diagnosis (p < .001). The cost reduction was $3412 (7.2%) if NIVH was begun 0–15 days after diagnosis (p < .001), and the cost of care benefit disappeared if NIVH was begun more than 15 days after diagnosis. Importantly, even in the hypercapnic group started more than 15 days following diagnosis, NIVH use was not associated with an increase in Medicare expenditures.
The unspecified COPD-CRF phenotype had no change in costs for those started at 0–7 days, a $1184 decrease if started between 8 and 15 days (p < .0001), and a $1366 increase in Medicare expenditures for those begun on NIVH between 16 and 30 days after diagnosis (p < .0001) resulting in cost neutrality for this group. In hypoxic COPD-CRF, NIVH use was associated with higher Medicare expenditures compared to controls in all treatment initiation windows (p < .001).
5.4 Device choice to deliver NIVH
Another area of controversy involves the particular devices best suited to deliver NIVH. Some experts recommend a step approach where less powerful and sophisticated, but cheaper, machines such as time-cycled bi-level devices termed respiratory assist devices by CMS (RADS, HCPCS EO471) or bi-level positive airway pressure devices (BPAP, HCPCS EO470) are used initially, with the more expensive home ventilators (HCPCS EO466) reserved for those who fail initial therapy [
[7]
,[9]
,[13]
]. To our knowledge, no clinical trial has been published testing this proposed “tried and failed” strategy.- Gay P.C.
- Owens R.L.
On behalf of the ONMAP technical expert panel, executive summary: optimal noninvasive ventilation Medicare access promotion (ONMAP).
Chest. 2021; https://doi.org/10.1016/j.chest.2021.05.074
Attempting to answer this question, we analyzed the data based on the type of device used to treat COPD-CRF. During the 4-year period from 2016 to 2020, total prescriptions for devices prior to exclusions showed 96 patients (1.2% of the NIVH cohort) received RADS, 512 (6.6%) received BPAP, and 7329 (92%) received home ventilators. The small sample sizes of RADS and BPAP patients did not allow for a statistically meaningful analysis of their effect on COPD-CRF outcomes. Therefore, our results are limited to patients treated with ventilators and our opinion is that given the overwhelming use of ventilators by US providers, these devices are the current standard of care to treat hypercapnic COPD-CRF. Use of RADS or BPAP in this population should be confined to clinical trials until they have been shown safe and effective in treating hypercapnic COPD-CRF.
5.5 Limitations
The major limitation of this study is its retrospective, non-randomized design. To address this, we minimized confounding factors by adjusting the models using numerous socio-demographic and clinical factors. We limited selection bias by employing IPTW propensity scoring techniques and accounted for immortal time bias by modeling each beneficiary in the treatment group as a control group member for the period between the diagnosis date and the treatment initiation date, and as a treatment group member thereafter.
Since the Medicare 100% RIF claims data does not include information from non-fee-for-service populations such as health maintaince organizations, Part C comprehensive Medicare Advantage plans, or for people not covered by Medicare, these results should not be generalized to those populations.
Currently, 3.4 million COPD patients are enrolled in traditional fee-for-service (FFS) Medicare. This represents 60% of all Medicare COPD patients and 23% of all U.S. COPD patients [
[19]
]. Since COPD-CRF, the most severe form of COPD, is more common with advancing age, and since older Medicare patients are more likely to enroll in fee-for-service Medicare, it is likely that the COPD-CRF patients which are the subject of this study represent a significantly larger proportion of the overall Medicare and general U.S. COPD population then these percentages imply [[20]
,[21]
].Finally, since the Medicare RIF does not contain data regarding compliance with NIVH or the specific pCO2 level above normal used to establish hypercapnia, we cannot comment as to any correlation between hours of NIVH use or degree of pCO2 elevation and the outcomes presented in this paper.
6. Conclusions
This retrospective cohort study suggests that treating hypercapnic COPD-CRF with NIVH is associated with significant reductions in mortality, hospitalizations, and total Medicare costs. Additionally, the sooner NIVH is begun following the diagnosis of hypercapnic COPD-CRF, the greater are these benefits. Patients with hypoxic COPD-CRF or unspecified COPD-CRF do not appear to benefit from NIVH.
Author confirmation
As submitting author, I attest that I have written consent from all authors to submit the manuscript, and that all authors accept complete responsibility for the contents of the manuscript entitled “Early Initiation of Non-Invasive Ventilation at Home Improves Survival and Reduces Healthcare Costs in COPD Patients with Chronic Hypercapnic Respiratory Failure.”
This manuscript has not been previously published. It is not currently under consideration elsewhere. The work reported here will not be submitted for publication elsewhere until a final decision has been made as to its acceptability by Respiratory Medicine.
Finally, I attest that the manuscript is independent and original work, and all conclusions are from the authors. At the time of its writing, Dr. Frazier was employed by VieMed which funded the study, and the other four authors were employed by Dobson DaVanzo & Associates, LLC. The authors whose names are listed immediately below certify that they have NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or nonfinancial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in this manuscript:
William D. Frazier, Joan E. DaVanzo, Allen Dobson
Steven Heath
Komi Mati
CRediT authorship contribution statement
William D. Frazier: Writing – review & editing, Writing – original draft, Study concept and design, clinical research, review of clinical findings, draft manuscript. Joan E. DaVanzo: Writing – review & editing, Writing – original draft, Heath: Study concept and design, research, review, draft manuscript. Allen Dobson: Writing – review & editing, Writing – original draft, Heath: Study concept and design, research, review, draft manuscript. Komi Mati: Writing – review & editing, Writing – original draft, Econometric methodology and programmed all analyses using 100% claims, draft and reviewed manuscript.
Appendix.
Fig. 1AFlow chart of patients who met inclusion/exclusion criteria for the study population.
Table 1APatient Characteristics by CRF Phenotype
References
- Mortality in the United States, 2016. vol. 293. NCHS Data Brief, 2017
- Chronic obstructive pulmonary disease: a view from the NHLBI.Am. J. Respir. Crit. Care Med. 2008; 178: 441-443
- Interventions to improve communication at hospital discharge and rates of readmission: a systematic review and meta-analysis.JAMA Netw. Open. 2021; 4 (Epub 2021 Aug 2)e2119346
- Risk trajectories of readmission and death in the first year after hospitalization for chronic obstructive pulmonary disease.Am. J. Respir. Crit. Care Med. 2018; 197: 1009-1017
- Predictors of mortality in patients with COPD and chronic respiratory failure: the Quality-of-Life Evaluation and Survival Study (QuESS): a three-year study.COPD. 2016; 13: 130-138https://doi.org/10.3109/15412555.2015.1067294
- Health-related quality of life measurement in patients with chronic respiratory failure.Respir. Investig. 2018; 56: 214-221https://doi.org/10.1016/j.resinv.2018.01.006(2018)1009–1017
- Long-term noninvasive ventilation in chronic stable hypercapnic chronic obstructive pulmonary disease. An official American thoracic society clinical practice guideline Am.J. Respir. Crit. Care Med. 2020; 202: e74-e87https://doi.org/10.1164/rccm.202006-2382ST
- Non-invasive ventilation at home improves survival and decreases healthcare utilization in Medicare beneficiaries with Chronic Obstructive Pulmonary Disease with chronic respiratory failure.Respir. Med. 2021; 177: 1-6
- Association of home noninvasive positive pressure ventilation with clinical outcomes in chronic obstructive pulmonary disease: a systematic review and meta-analysis.JAMA. 2020; 323: 455-465https://doi.org/10.1001/jama.2019.22343
- Noninvasive ventilation at home reduces mortality in COPD with CRF.Am. J. Manag. Care. 2021; 27: e308-e315https://doi.org/10.37765/ajmc.2021.88743
- Non-invasive positive pressure ventilation for the treatment of severe stable COPD: a prospective, multicenter, randomized controlled trial.Lancet Respir. Med. 2014; 2: 698-705
- Effect of home NIV with oxygen vs. oxygen alone on hospital readmission or death after an acute COPD exacerbation.JAMA. 2017; 317 (10.1001.2017.4451): 2177-2186
- On behalf of the ONMAP technical expert panel, executive summary: optimal noninvasive ventilation Medicare access promotion (ONMAP).Chest. 2021; https://doi.org/10.1016/j.chest.2021.05.074
- Non-invasive home ventilation in patients over 75 years of age: tolerance, compliance, and impact on quality of life.Respir. Med. 1998; 92: 1311-1320https://doi.org/10.1016/s0954-6111(98)90135-4
- Outcomes in patients with chronic obstructive pulmonary disease and obstructive sleep apnea: the overlap syndrome.Am. J. Respir. Crit. Care Med. 2010; 182: 325-331https://doi.org/10.1164/rccm.200912-1869OC
- An introduction to propensity score methods for reducing confounding in observational studies.Multivariate Behav. Res. 2011; 46: 399-424
- Reducing bias in observational studies using subclassification on the propensity score.J. Am. Stat. Assoc. 1984; 79: 516-524
- Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples.Stat. Med. 2009; 8: 3083-3107
- Medicare Enrollees with COPD Compared to the General Population.Avalere Health, 2021: 1-6
CDC.gov.COPD (accessed June 3, 2022).
CDC.gov> NCHS> COPD Publications and Information (accessed June 3, 2022).
Article info
Publication history
Published online: June 29, 2022
Accepted:
June 12,
2022
Received in revised form:
June 9,
2022
Received:
March 29,
2022
Identification
Copyright
© 2022 The Authors. Published by Elsevier Ltd.
User license
Creative Commons Attribution (CC BY 4.0) | How you can reuse 
Elsevier's open access license policy
Creative Commons Attribution (CC BY 4.0)
Permitted
- Read, print & download
- Redistribute or republish the final article
- Text & data mine
- Translate the article
- Reuse portions or extracts from the article in other works
- Sell or re-use for commercial purposes
Elsevier's open access license policy
00185-8&id=fx1.jpg){kind=link}