Improved medication adherence in COPD patients using tiotropium or tiotropium olodaterol with the HealthPrize digital behavior change program

ABSTRACT

Objective

To assess the impact of the HealthPrize RespiPoints™ program on treatment adherence and persistence in adults with chronic obstructive pulmonary disease (COPD).

Methods

In this retrospective cohort study, program participants and nonparticipants receiving tiotropium bromide (TIO) or TIO and olodaterol between 1 January 2015–31 March 2020 were propensity score matched (PSM), from the linked database of the HealthPrize patient list and IQVIA PharMetrics® Plus. Treatment adherence, persistence, healthcare resource utilization, and costs were compared. Multivariable logistic regression models assessed the odds of adherence (≥80% proportion of days covered [PDC]), adjusted risk of discontinuation, and adjusted total healthcare costs.

Results

Program participants (n = 262) demonstrated a 44% greater adherence during followup than nonparticipants (n = 262) (mean [standard deviation] PDC: 0.72 [0.27] vs 0.50 [0.36], p < 0.0001). Participants had higher odds of adherence vs nonparticipants (adjusted odds ratio: 2.51; 95% confidence interval: 1.72–3.66, p < 0.0001) and a lower percentage of participants discontinued their index medication (19.85% vs 33.59%, p = 0.0004). Fewer participants were hospitalized during follow-up (13.74% vs 17.56%, p = 0.23); adjusted total medical costs were 24% lower (p = 0.08). Higher pharmacy costs partially offset lower healthcare costs.

Conclusions

Program participants showed improved COPD medication adherence and persistence compared to nonparticipants.

KEYWORDS:

• Chronic obstructive pulmonary disease
• digital health
• healthcare resource utilization
• medication adherence
• behavior change
• real-world data

1. Introduction

Nonadherence to chronic obstructive pulmonary disease (COPD) medications significantly compromises clinical outcomes, including increased exacerbations, hospitalizations, quality of life, and mortality [1–3]. Adherence rates to prescribed pharmacotherapy typically range from 20% to 60% [3–8]. with many patients filling prescriptions just once before discontinuing [9]. Consequently, treatment effectiveness is often far lower than that achieved in clinical trials [1,4,10]. Improving adherence, therefore, is of central importance in COPD management.

Improved medication adherence results in significantly fewer hospitalizations and reduced healthcare costs [11–16]. Because of the complexity of the nonadherence problem, including the wide diversity of factors that contribute to nonadherence, improving adherence to COPD treatment remains a significant challenge [3,5,6,10]. Numerous interventions including digital programs have attempted to improve adherence [5,17,18]. Most efforts, however, fail to have a meaningful impact [2,8,9].

RespiPoints™ is a digital patient adherence program developed through the partnership of HealthPrize and Boehringer Ingelheim. It was designed to promote healthy behaviors and better management of COPD, including long-term treatment adherence, specifically for patients prescribed tiotropium bromide (TIO; Spiriva® Respimat®) or TIO and olodaterol (TIO/OLO; Stiolto® Respimat®). The program combines insights from behavioral science, medical education, and gamification techniques to harness both extrinsic and intrinsic motivation and promote frequent engagement. This study evaluated the impact of the program on treatment adherence and persistence over 12 months against matched controls. Secondary objectives evaluated the impact of the program on healthcare resource utilization (HCRU) and costs.

2. Patients and methods

2.1. Study design and data sources

A retrospective, observational, cohort study was conducted among adults with COPD (aged ≥18 years) receiving TIO or TIO/OLO in United States (US) clinical practice. Two cohorts were identified based on self-enrollment participation (or nonparticipation) in the program. Self-enrollment was available through multiple channels, including physician invitation and existing patient support resources, such as customer relationship management programs, co-pay support programs, and treatment websites. The program uses a proprietary gamification approach and patented technology combined with behavioral science and patient psychology strategies to facilitate the promotion of healthy behaviors, including better management of COPD and tracking of medication adherence among COPD patients treated with TIO or TIO/OLO (Supplementary Figure S1). Participants were identified using a linked database derived from the HealthPrize patient list and the IQVIA PharMetrics® Plus database. The patient list contained deidentified patient-level information on program participants, including program join date, treatments administered, and program engagement. These data were linked in a Health Insurance Portability and Accountability Act (HIPAA)-compliant manner to the IQVIA database. A comparison group of adults with COPD receiving TIO or TIO/OLO who were not enrolled in the patient list (nonparticipants) was identified in the IQVIA database. The IQVIA database includes historical information on patient demographics and fully adjudicated medical and pharmacy claims since 2006; the data are anonymized and HIPAA-compliant to protect patient privacy. The study design is presented in Figure 1.

Figure 1. Study design schematic.

Linked data from the patient list and IQVIA PharMetrics® Plus were utilized, with data from January 1, 2015, to March 31, 2020 (study period), for the program participant group. The index date was defined as the first such prescription of TIO or TIO/OLO for the participant group. To obtain a comparison group of nonparticipants using TIO or TIO/OLO, IQVIA’s PharMetrics® Plus database was used, which excluded patients who had linkage to the patient list during the study period. The index date for non-participants was selected as a random date within the selection window, which was assigned based on the specific month/year distribution of index dates of the participant group.HCRU, healthcare resource utilization; OLO, olodaterol; TIO, tiotropium bromide.

2.2. Study population

The index date for eligible participants was ≥ 1 prescription claims for TIO or TIO/OLO during the 30 days before or following enrollment between 1 January 2016, and 31 March 2019 (selection window). Nonparticipants had ≥ 1 prescription of TIO or TIO/OLO within the selection window, defined as the index date based on the participant cohort’s month/year distribution of index dates. All eligible patients had to be ≥18 years at the index date (first prescription of TIO or TIO/OLO), have continuous enrollment in the IQVIA database for ≥12 months in the pre- and post-index periods, and ≥ 1 COPD diagnosis code either International Classification of Diseases 9 (ICD-9) (496.xx) or ICD-10 (J44) on a confirmatory claim in the 12-month pre-index period (Figure 1). Exclusion criteria included ≥ 1 asthma diagnosis code either ICD-9 (493.xx) or ICD-10 (J45) on a confirmatory claim in the 12-month pre-index period, incomplete data or data quality issues (invalid or missing year of birth, sex, region, or health plan enrollment dates), Medicare or State Children’s Health Insurance Program coverage, and those aged ≥65 years at the index date who were not covered by Medicare Risk.

2.2.1. Treatment-experienced patients

After applying the exclusion criteria (Figure 2), 95.97% of prematched participants and 73.12% of prematched nonparticipants were found to be treatment-experienced. Treatment experience was defined as ≥ 1 prescription claims for the following COPD maintenance medications over the pre-index period (until 30 days before the index date): long-acting muscarinic antagonists (LAMAs), long-acting β₂-agonists (LABAs), LAMA/LABA combination, inhaled corticosteroids (ICS)/LABA combination, ICS/LAMA/LABA combination, phosphodiesterase-4 (PDE-4) inhibitors, and methylxanthines. Patients new to COPD maintenance medications (treatment naïve) were excluded from the analysis. These patients were matched using propensity score matching (PSM) technique as described in section 2.4. Statistical analyses).

Figure 2. Flow diagram of patient disposition.

After selecting patients aged ≥18 years at the index date, all program participants and 99.13% of the non-participant initial groups remained in the study. After selecting participants with at least 12 months of continuous enrollment (CE) in both the pre-index and post-index periods, 46.60% of the initial participants and 38.18% of the nonparticipants remained in the study. After selecting patients with ≥ 1 ICD-9 or ICD10 diagnosis code of COPD in the pre-index period, 38.35% and 26.50% of the initial participants and nonparticipants, respectively, remained in the study. After applying the exclusion criteria, i.e. linkage to HealthPrize patient list, age ≥65 years at the index date, and no coverage under Medicare Risk due to incomplete Medicare fee-for-service data or data quality issues, 24.75% and 14.54% of the initial participants and nonparticipants, respectively, remained in the study. After applying the exclusion criteria, 95.97% (262 of 273) and 73.12% (24,036 of 32,872) of prematched program participants and nonparticipants, respectively, were found to be treatment-experienced patients (≥1 prescription claim of COPD maintenance medication in the 12- to 1-month pre-index period). After selecting patients who were considered treatment-experienced, the final sample comprised 262 (23.75%) and 24,036 (10.63%) patients in the program participant and nonparticipant groups, respectively.CE, continuous enrollment; COPD, chronic obstructive pulmonary disease; ICD-9/10, International Classification of Diseases-9/10.

2.3. Study variables and outcomes

2.3.1. Treatment adherence, persistence, switching, and augmentation

Outcomes included adherence, persistence, medication switching, and augmentation. Patients were followed up until they met the outcome of interest or until they were censored (last date of continuous eligibility in the database or end of the study data period [12 months post index]), whichever occurred first. Pre- and postmatching baseline characteristics were evaluated between participants and nonparticipants. Adherence to the index treatment (TIO or TIO/OLO) was evaluated and compared for matched participants and nonparticipants. Adherence was evaluated as the proportion of days covered (PDC) over the 12-month post-index period (calculated as the number of days/year with drugs on hand, divided by 365 days). The numerator of the PDC was adjusted to avoid double counting of days when the patient had multiple fills on hand. PDC was reported as a continuous variable (mean, standard deviation [SD], median) and by quintiles. The gap days between fills after the index date in the post-index period were reported as a continuous variable.

Persistence was calculated based on the time (in consecutive days) from index drug initiation until discontinuation, medication switch, augmentation, or end of the study period, whichever occurred first. Discontinuation was defined as a gap in a series of successive index prescriptions ≥ 2 times the duration of the last prescription in that series, before the gap, measured from the last refill. Patients were considered to have switched if they filled a non-index COPD medication <30 days since index discontinuation and had ≥ 30-day supply of switched-to medication in the follow-up period. Augmentation was defined as ≥ 30-day supply of an additional non-index COPD medication with ≥30 days of overlapping use with index medication. Persistence with index COPD treatment for ≥30 days following the second medication differentiated augmentation from switching. The number and proportion of patients in each group who persisted with treatment, discontinued, switched non-index treatment, or augmented treatment were evaluated, including mean time from index medication to switch or discontinuation (mean [SD], or median).

2.3.2. All-cause and COPD-specific HCRU and costs

COPD-specific HCRU included claims with a COPD diagnosis code in the primary (first) position for inpatient claims, any position for outpatient claims, and/or therapies used in the treatment of COPD. HCRU and costs were calculated on a per-patient basis, averaged across the group (unless otherwise specified). Direct healthcare costs were determined using the ‘allowed’ amount field, which represented the reimbursed amount paid by payers before adjusting for out-of-pocket costs and coordination of benefits. Costs were converted to 2020 US dollar (USD) rates using the medical component of the Consumer Price Index [19]. Cost outcomes included total healthcare costs, pharmacy costs, inpatient hospitalizations, and total outpatient costs.

2.4. Statistical analyses

Baseline demographic and clinical characteristics were assessed for participants and nonparticipants before and after propensity score matching (PSM) to adjust for treatment selection bias. Demographics were from the index date; clinical characteristics were measured over the 12-month (360-day) pre-index period (not including the index date, unless otherwise specified).

Stratified PSM was performed using a logistic regression model and a 1:1 match was conducted for the pairwise comparison of program enrollment (Yes vs No). Differences were evaluated based on an absolute standardized mean difference (SMD) ≥0.10. Unsuccessfully matched patients were excluded. Continuous variables were presented as mean (SD) and median values where relevant. Reporting mean (SD) was favored due to the presence of values close to zero in various parameters when the median value was used.

Bivariate comparisons were made using the chi-square test for categorical variables. For comparing continuous variables between the two groups, an independent sample t test (mean) or Wilcoxon ranksum test (median) was used. All tests were conducted assuming a two-tailed test of significance, and a p value of < 0.05 was considered statistically significant. Time-to-event (index treatment discontinuation or switch, whichever occurred first) was measured using the Kaplan-Meier (KM) method. A multivariable logistic regression model was used to examine the odds of being fully adherent (≥80% PDC) for participants vs nonparticipants, adjusting for potential confounders. A multivariable Cox proportional hazards model was used to measure the adjusted risk of discontinuation for participants vs nonparticipants. One generalized linear model (GLM) with log-link/gamma distribution was used to estimate the adjusted total healthcare costs for participants vs nonparticipants, where adherence was used as an independent variable. Covariates considered for inclusion in the models had postmatching SMD ≥ 0.10 after checking for multicollinearity. All analyses were based on observed, not projected, data. Analyses were conducted using SAS version 9.4 (SAS, Cary, NC, U.S.A.).

3. Results

3.1. Study population

The study initially identified 1,103 participants (Figure 2). In the final sample, 262 participants and 24,036 nonparticipants were considered in the prematched group. The postmatched group included 262 participants and 262 nonparticipants.

3.2. Baseline characteristics

Pre- and postmatching baseline characteristics are presented in Tables 1 and 2. Prematched characteristics were generally imbalanced between groups (SMD ≥0.10) and therefore, PSM variables included age (continuous), gender, region, payer type, index year, Elixhauser score (continuous, using composite score instead of individual comorbidities), chronic pain/fibromyalgia, medications of interest (index prescription [TIO, TIO/OLO] and number of COPD medication claims), co-pay amount of index medication, and time from last COPD diagnosis to index date. Postmatching characteristics were generally well balanced. Preventive services (influenza and pneumococcal vaccinations and screening tests such as mammograms and prostate-specific antigen tests) were not included in the PSM. Based on remaining imbalances post PSM, the following variables were adjusted for in the multivariate models: Elixhauser Comorbidity Index, vaccine administration, ICS/LABA use, index medication co-pay, patients with ≥ 1 hospitalization, and total pre-index healthcare costs. Following adjustment, the Elixhauser scores showed balanced SMDs, while index prescription ICS/LABA and influenza and pneumococcal vaccines remained imbalanced. Imbalances remained for index medication co-pay amount (<$50 and ≥$50), patients having ≥ 1 hospitalization, and total healthcare costs per patient. Some clinical variables remained imbalanced, for example, cardiac valvular disease, hypertension, and smoking.

Table 1. Baseline demographic and clinical characteristics for program participants and nonparticipants before and after PSM.

	Prematching					Postmatching
	Program participants		Nonparticipants			Program participants		Nonparticipants
	N = 262		N = 24,036			N = 262		N = 262
Demographic characteristics*	n	%	n	%	SMD^†	n	%	n	%	SMD^†
Age (years)
Mean	58.9		62.6		0.41	58.9		58.7		−0.07
SD	5.5		9.4		0.41	5.5		5.4
Median	59		61		0.41	59		59
Age group (years), n (%)
18–34	0	0.00%	61	0.25%	0.07	0	0.00%	0	0.00%	0.00
35–44	3	1.15%	335	1.39%	0.02	3	1.15%	2	0.76%	−0.04
45–54	40	15.27%	3,103	12.91%	−0.07	40	15.27%	49	18.70%	0.09
55–64	203	77.48%	14,118	58.74%	−0.41	203	77.48%	198	75.57%	−0.05
≥65	16	6.11%	6,419	26.71%	0.58	16	6.11%	13	4.96%	−0.05
Sex, n (%)
Female	151	57.63%	12,600	52.42%	−0.10	151	57.63%	158	60.31%	0.05
Male	111	42.37%	11,436	47.58%	0.1	111	42.37%	104	39.69%	−0.05
Geographic region, n (%)
Northeast	31	11.83%	5,047	21.00%	0.25	31	11.83%	24	9.16%	−0.09
Midwest	90	34.35%	8,284	34.46%	0	90	34.35%	95	36.26%	0.04
South	123	46.95%	8,589	35.73%	−0.23	123	46.95%	135	51.53%	0.09
West	18	6.87%	2,116	8.80%	0.07	18	6.87%	8	3.05%	−0.18
Payer type, n (%)
Commercial	163	62.21%	10,089	41.97%	−0.41	163	62.21%	163	62.21%	0.00
Medicaid	0	0.00%	543	2.26%	0.22	0	0.00%	1	0.38%	0.09
Medicare risk	18	6.87%	7,716	32.10%	0.67	18	6.87%	16	6.11%	−0.03
Self-insured	81	30.92%	5,652	23.51%	−0.17	81	30.92%	82	31.30%	0.01
Other/unknown	0	0.00%	36	0.15%	0.05	0	0.00%	0	0.00%	0.00
Health plan type, n (%)
Consumer-directed healthcare	4	1.53%	220	0.92%	−0.06	4	1.53%	2	0.76%	−0.07
HMO	41	15.65%	7,398	30.78%	0.36	41	15.65%	30	11.45%	−0.12
Indemnity	1	0.38%	169	0.70%	0.04	1	0.38%	1	0.38%	0.00
POS	18	6.87%	908	3.78%	−0.14	18	6.87%	15	5.73%	−0.05
PPO	198	75.57%	15,303	63.67%	−0.26	198	75.57%	214	81.68%	0.15
Other/unknown	0	0.00%	38	0.16%	0.06	0	0.00%	0	0.00%	0.00
Index year, n (%)
2016	44	16.79%	6,523	27.14%	0.25	44	16.79%	49	18.70%	0.05
2017	84	32.06%	7,719	32.11%	0	84	32.06%	79	30.15%	−0.04
2018	93	35.50%	7,809	32.49%	−0.06	93	35.50%	99	37.79%	0.05
2019	41	15.65%	1,985	8.26%	−0.23	41	15.65%	35	13.36%	−0.07
Program verified, n (%)
Yes	167	63.74%	0	0.00%	−1.88	167	63.74%	0	0.00%	−1.88
No	95	36.26%	24,036	100.00%	1.88	95	36.26%	262	100.00%	1.88

Notes: Demographic characteristics that were well balanced between cohorts (absolute SMD < 0.1) after matching are shown in bold.

*Assessed as of the patient index date.

*For continuous variables, SMD was calculated as: $d=\frac{\left(\overline{X_1}-\overline{X_2}\right)}{\sqrt{\frac{s_1^2+s_2^2}{2}}}$^†For categorical variables, SMD was calculated as: $d=\frac{\left(p_1-p_2\right)}{\sqrt{\frac{p_1\left(1-p_2\right)+p_2\left(1-p_1\right)}{2}}}$ where $\bar{X}$ and p denote the sample means and proportions in the groups, respectively, and s denotes sample variances.

Table 2. Baseline clinical characteristics for program participants and nonparticipants before and after PSM.

	Prematching					Postmatching
	Program participants		Nonparticipants			Program participants		Nonparticipants
	N = 262		N = 24,036			N = 262		N = 262
Clinical characteristics^§	n	%	n	%	SMD^†	n	%	n	%	SMD^†
Physician specialty at index, n (%)
Primary care physician	104	39.69%	11,129	46.30%	0.13	104	39.69%	102	38.93%	−0.02
Pulmonologist	38	14.50%	3,484	14.49%	0	38	14.50%	31	11.83%	−0.08
Others	115	43.89%	8,677	36.10%	−0.16	115	43.89%	122	46.56%	0.05
Unknown	5	1.91%	746	3.10%	0.08	5	1.91%	7	2.67%	0.05
CCI score, n (%)
0	0	0.00%	134	0.56%	0.11	0	0.00%	1	0.38%	0.09
1	135	51.53%	10,091	41.98%	−0.19	135	51.53%	146	55.73%	0.08
2	57	21.76%	4,581	19.06%	−0.07	57	21.76%	46	17.56%	−0.11
3	38	14.50%	3,637	15.13%	0.02	38	14.50%	36	13.74%	−0.02
≥4	32	12.21%	5,593	23.27%	0.29	32	12.21%	33	12.60%	0.01
Mean (SD)	2	(1.5)	2.6	(2)	0.27	2	(1.5)	2	(1.6)	−0.07
Median	1		2			1		1
Comorbidities of interest
Alcohol/drug abuse	12	4.58%	2,213	9.21%	0.18	12	4.58%	22	8.40%	0.16
Cardiac arrhythmia	27	10.31%	3,835	15.96%	0.17	27	10.31%	27	10.31%	0.00
Cardiac valvular disease	26	9.92%	2,633	10.95%	0.03	26	9.92%	17	6.49%	−0.13
Cerebrovascular disease	17	6.49%	2,335	9.71%	0.12	17	6.49%	16	6.11%	−0.02
Chronic kidney disease	9	3.44%	2,180	9.07%	0.23	9	3.44%	13	4.96%	0.08
Chronic pain/fibromyalgia	32	12.21%	3,799	15.81%	0.1	32	12.21%	36	13.74%	0.05
Congestive heart failure	19	7.25%	3,430	14.27%	0.23	19	7.25%	23	8.78%	0.06
Depression	40	15.27%	4,985	20.74%	0.14	40	15.27%	46	17.56%	0.06
Diabetes mellitus	44	16.79%	5,596	23.28%	0.16	44	16.79%	48	18.32%	0.04
Dyslipidemia	129	49.24%	13,730	57.12%	0.16	129	49.24%	127	48.47%	−0.02
Hypertension	157	59.92%	15,784	65.67%	0.12	157	59.92%	141	53.82%	−0.12
Liver/GB/pancreatic disease	20	7.63%	1,720	7.16%	−0.02	20	7.63%	15	5.73%	−0.08
Myocardial infarction/CAD	41	15.65%	5,866	24.41%	0.22	41	15.65%	39	14.89%	−0.02
Osteoarthritis	63	24.05%	6,944	28.89%	0.11	63	24.05%	67	25.57%	0.04
Peripheral vascular disease	32	12.21%	3,858	16.05%	0.11	32	12.21%	24	9.16%	−0.10
Renal failure/dialysis	9	3.44%	2,180	9.07%	0.23	9	3.44%	13	4.96%	0.08
Sleep disorders	80	30.53%	7,001	29.13%	−0.03	80	30.53%	66	25.19%	−0.12
Smoking	160	61.07%	15,258	63.48%	0.05	160	61.07%	176	67.18%	0.13
Thyroid disease	44	16.79%	4,429	18.43%	0.04	44	16.79%	36	13.74%	−0.09
Elixhauser comorbidity index, n (%)
0	0	0.00%	71	0.30%	0.08	0	0.00%	1	0.38%	0.09
1	38	14.50%	2,995	12.46%	−0.06	38	14.50%	48	18.32%	0.10
2	75	28.63%	4,489	18.68%	−0.24	75	28.63%	67	25.57%	−0.07
3	58	22.14%	4,288	17.84%	−0.11	58	22.14%	50	19.08%	−0.08
≥4	91	34.73%	12,193	50.73%	0.33	91	34.73%	96	36.64%	0.04
Mean (SD)	3.4	(2.2)	4.2	(2.8)	0.33	3.4	(2.2)	3.3	(2.2)	−0.03
Median	3		4			3		3
Medications of interest, n (%)
Index prescription (on index date)
TIO	192	73.28%	21,338	88.78%	0.4	192	73.28%	199	75.95%	0.06
TIO/OLO	70	26.72%	2,698	11.22%	−0.40	70	26.72%	63	24.05%	−0.06
COPD-related
Antibiotics	168	64.12%	14,867	61.85%	−0.05	168	64.12%	164	62.60%	−0.03
Oral corticosteroids	148	56.49%	12,603	52.43%	−0.08	148	56.49%	140	53.44%	−0.06
ICS	25	9.54%	1,687	7.02%	−0.09	25	9.54%	22	8.40%	−0.04
SABA	190	72.52%	16,835	70.04%	−0.05	190	72.52%	186	70.99%	−0.03
SAMA	8	3.05%	617	2.57%	−0.03	8	3.05%	8	3.05%	0.00
LABA	6	2.29%	433	1.80%	−0.03	6	2.29%	7	2.67%	0.02
LAMA	192	73.28%	18,978	78.96%	0.13	192	73.28%	192	73.28%	0.00
LAMA/LABA	67	25.57%	2,872	11.95%	−0.35	67	25.57%	57	21.76%	−0.09
LAMA (excluding TIO)	12	4.58%	793	3.30%	−0.07	12	4.58%	19	7.25%	0.11
LAMA/LABA (excluding TIO/OLO)	8	3.05%	766	3.19%	0.01	8	3.05%	8	3.05%	0.00
ICS/LABA	140	53.44%	13,872	57.71%	0.09	140	53.44%	155	59.16%	0.12
ICS/LAMA/LABA	3	1.15%	71	0.30%	−0.10	3	1.15%	1	0.38%	−0.09
Phosphodiesterase-4 inhibitors	10	3.82%	730	3.04%	−0.04	10	3.82%	12	4.58%	0.04
Methylxanthines	6	2.29%	457	1.90%	−0.03	6	2.29%	4	1.53%	−0.06
β2-Agonist other	1	0.38%	65	0.27%	−0.02	1	0.38%	0	0.00%	−0.09
SAMA/SABA	43	16.41%	4,164	17.32%	0.02	43	16.41%	46	17.56%	0.03
Mean (SD) number of COPD medication claims	18	(11.4)	16.1	(12.1)	−0.25	18	(11.4)	19.5	(16)	−0.04
Median	15		13			15		14
Preventive services, n (%)
Vaccines
Influenza	147	56.11%	12,491	51.97%	−0.08	147	56.11%	114	43.51%	−0.25
Pneumococcal	48	18.32%	4,049	16.85%	−0.04	48	18.32%	30	11.45%	−0.19
Screening test
Mammogram	63	24.05%	4,915	20.45%	−0.09	63	24.05%	54	20.61%	−0.08
PSA test	46	17.56%	3,341	13.90%	−0.10	46	17.56%	38	14.50%	−0.08
Co-pay amount for index medication, n (%)
<$50	121	46.18%	15,694	65.29%	0.39	121	46.18%	159	60.69%	0.29
≥$50	141	53.82%	8,342	34.71%	−0.39	141	53.82%	103	39.31%	−0.29
Mean (SD)	110.5	(174)	60.2	(92.2)	−0.46	110.5	(174)	101.5	(178.6)	−0.26
Median	53		38			53		43
Days’ supply of index treatment
30	215	82.06%	18,841	78.39%	−0.09	215	82.06%	213	81.30%	−0.02
90	41	15.65%	4,693	19.52%	0.1	41	15.65%	45	17.18%	0.04
Others	6	2.29%	502	2.09%	−0.01	6	2.29%	4	1.53%	−0.06
Mean (SD)	39.5	(21.9)	42.1	(24)	0.14	39.5	(21.9)	40.6	(22.7)	0.09
Median	30		30			30		30
Mean (SD) time from last COPD diagnosis to index date (days)	77.2	(82.5)	65.9	(82.6)	−0.23	77.2	(82.5)	74.4	(84.1)	−0.08
Median	50		29			50		44
Mean (SD) time from program enrollment to index date (days)	6.5	(61.5)				6.5	(61.5)
Median	−6					−6
<0	164	62.60%				164	62.60%
0–30	69	26.34%				69	26.34%
31–60	13	4.96%				13	4.96%
61–90	7	2.67%				7	2.67%
>90	9	3.44%				9	3.44%
Patients with ≥ 1 prescription fill	262	100.00%	24,036	100.00%	0	262	100.00%	262	100.00%	0.00
Number of prescriptions based on GPI-8
0–1	1	0.38%	60	0.25%	−0.02	1	0.38%	1	0.38%	0.00
2–3	9	3.44%	600	2.50%	−0.06	9	3.44%	6	2.29%	−0.07
3–4	9	3.44%	683	2.84%	−0.03	9	3.44%	8	3.05%	−0.02
≥5	243	92.75%	22,693	94.41%	0.07	243	92.75%	247	94.27%	0.06
Mean (SD) total pharmacy costs per patient (USD)	11,590.30	(21,481.40)	10,901.60	(23,041.40)	−0.07	11,590.30	(21,481.40)	13,077.70	(31,266.70)	0.02
Median	6,409.00		6,076.00			6,409.00		6,631.00
Patients with ≥ 1 outpatient visit	262	100.00%	24,036	100.00%	0	262	100.00%	262	100.00%	0.00
Mean (SD) total outpatient costs per patient (USD)	30,202.00	(47,127.00)	31,217.90	(53,057.10)	0.07	30,202.00	(47,127.00)	34,149.10	(64,968.40)	0.09
Median	18,448.00		19,038.00			18,448.00		18,782.00
Patients with ≥ 1 hospitalization	34	12.98%	5,597	23.29%	0.27	34	12.98%	54	20.61%	0.21
Mean (SD) total inpatient costs per patient (USD)	4,923.00	(18,269.10)	8,213.70	(27,515.50)	0.26	4,923.00	(18,269.10)	8,428.40	(23,004.00)	0.21
Mean (SD) total healthcare costs per patient (USD)	23,534.70	(34,041.50)	28,530.00	(46,027.20)	0.19	23,534.70	(34,041.50)	29,499.80	(42,887.50)	0.18
Median	12,131.00		14,311.00			12,131.00		13,217.00

Clinical characteristics that were well balanced between cohorts (absolute SMD < 0.1) after matching are shown in bold.

^§Clinical characteristics were assessed over the 6-month pre-index period, not including the index date^.*For continuous variables, SMD was calculated as: $d=\frac{\left(\overline{X_1\hspace{0.17em}}-\hspace{0.17em}\overline{X_2}\right)}{\sqrt{\frac{s_1^2\hspace{0.17em}+\hspace{0.17em}{s}_2^2}{2}}}$^†For categorical variables, SMD was calculated as: $d=\hspace{0.17em}\frac{\left(p_1-\hspace{0.17em}{p}_2\right)}{\sqrt{\frac{p_1\left(1-p_2\right)+\hspace{0.17em}{p}_2\left(1-\hspace{0.17em}{p}_1\right)}{2}}}$.

where $\bar{X}$ and p denote the sample means and proportions in the groups, respectively, and s denotes sample variances.

Abbreviations: AIDS, acquired immunodeficiency syndrome; CAD, coronary artery disease; CCI, Charlson Comorbidity Index; COPD, chronic obstructive pulmonary disease; GB, gall bladder; GPI-8, Generic Product Identifier-8; HIV, human immunodeficiency virus; HMO, health maintenance organization; ICS, inhaled corticosteroids; LABA, long-acting β₂-agonist; LAMA, long-acting muscarinic antagonist; OLO, olodaterol; POS, point-of-service; PPO, preferred provider organization; PSA, prostate-specific antigen; PSM, propensity score matching; SABA, short-acting β₂-agonist; SAMA, short-acting muscarinic antagonist; SD, standard deviation; SMD, standardized mean difference; TIO, tiotropium bromide; USD, United States dollar.

3.3. Adherence

After adjusting for the imbalanced variables, participants had significantly greater adherence over the 12-month post-index period vs nonparticipants, with a mean [SD] PDC of 0.72 [0.27] vs 0.50 [0.36] (44%, p < 0.0001; Table 3). The proportion of participants considered fully adherent (≥80% PDC) was significantly higher vs that of nonparticipants (53.82% vs 33.97%, p < 0.0001). The number of index prescription fills was significantly higher for participants vs nonparticipants (mean [SD]: 8.28 [3.89] vs 5.47 [4.38]; median: 9.00 vs 4.50; p < 0.0001). The number of gap days between index fills was also lower for participants vs nonparticipants (mean [SD]: 94.07 [101.00] vs 177.35 [133.74]; median: 53.50 vs 185.00; p < 0.0001; Table 3). After adjusting for key covariates, multivariable logistic regression revealed that participants had 2.5 times higher odds of being fully adherent to the index medication vs nonparticipants (adjusted odds ratio: 2.51; 95% confidence interval [CI]: 1.72–3.66; p < 0.0001; Table 4).

Table 3. Treatment adherence and persistence for program participants vs nonparticipants (postmatching).

	Program participants					Nonparticipants					p value
	n = 262					n = 262
Adherence measures*	n	%	Mean	SD	Median	n	%	Mean	SD	Median	Mean or %	Median
Adherence^†
PDC over follow-up			0.72	0.27	0.83			0.50	0.36	0.49	<0.0001	<0.0001
Q1 (≤20%)	19	7.25%				85	32.44%				<0.0001
Q2 (20%–40%)	19	7.25%				35	13.36%				0.0215
Q3 (40%–60%)	36	13.74%				26	9.92%				0.1762
Q4 (60%–80%)	47	17.94%				27	10.31%				0.0121
Q5 (≥80%)	141	53.82%				89	33.97%				<0.0001
Mean number of index prescription fills			8.28	3.89	9.00			5.47	4.38	4.50	<0.0001	<0.0001
Mean gap days between fills			94.07	101.00	53.50			177.35	133.74	185.00	<0.0001	<0.0001
Patients discontinuing their index medication^‡	52	19.85%				88	33.59%				0.0004
Patients switching their index medication^§	69	26.34%				83	31.68%				0.1777
Patients who switched from index medication to any other COPD medication	63	24.05%				82	31.30%				0.0636
Time from index medication to switch			113.86	85.27	90.00			69.95	83.53	38.00	0.0017	0.0002
Patients who switched from TIO to TIO/OLO	5	1.91%				0	0.00%
Patients who switched from TIO/OLO to TIO	1	0.38%				1	0.38%
Patients who augmented their index medication^¶	107	40.84%				76	29.01%				0.0045
Patients persistent on their index medication	34	12.98%				15	5.73%				0.0044
Number of persistent days^‖			251.08	136.16	365.00			169.23	155.43	90.00	<0.0001

Notes: *Assessed over the 12-month post-index period among the subset of patients with 12-month continuous enrollment, including the index date.

†Adherence computed as PDC over the 12-month post-index period, calculated as the number of days with drug on hand during the 365-day period, divided by 365.

*Defined as the occurrence of a gap in a series of successive prescriptions for the index drug that was greater than or equal to two times the duration of the last prescription in that series, before the gap, measured from the last refill of that medication.

^†Patients were considered to have switched if they filled a nonindex COPD treatment <30 days since index discontinuation, and patients were required to have at least a 30-day supply of the switched-to drug in the follow-up period.

^‡Augmentation was defined as the addition of a nonindex COPD medication to the index medication while still persistent to the index medication (if even appropriate). Patients were required to have at least 30 days’ supply of the augmentation medication and at least 30 days of overlapping use with their index medication. The continuing use of index COPD treatment following the second medication for ≥30 days differentiated augmentation from switching.

^§Persistence was calculated based on the time (in consecutive days) from index drug initiation until discontinuation, medication switch, augmentation, or end of the study period (12 months after index), whichever occurred first.

Comparisons were conducted using a parametric t test (mean) and nonparametric Wilcoxon rank-sum test (median) and chi-square test for categorical variables.

Abbreviations: COPD, chronic obstructive pulmonary disease; OLO, olodaterol; PDC, proportion of days covered; Q, quintile; SD, standard deviation; TIO, tiotropium bromide.

Table 4. Logistic regression analysis for adherence.

Variable	Odds ratio (95% CI)	p value
Program enrollment (key variable)
Program participants (vs nonparticipants)	2.51 (1.72–3.66)	<0.0001
Comorbidities
Smoking	1.35 (0.91–1.99)	0.137
Sleep disorders	0.89 (0.58–1.36)	0.583
Elixhauser score
2 (vs 0–1)	0.40 (0.23–0.72)	0.002
3 (vs 0–1)	0.52 (0.28–0.95)	0.032
≥4 (vs 0–1)	0.28 (0.15–0.52)	<0.0001
ICS/LABA use (pre-index)	0.80 (0.55–1.16)	0.234
Influenza vaccine (pre-index)	1.46 (1.00–2.11)	0.048
Pneumococcal vaccine (pre-index)	1.14 (0.68–1.91)	0.630
Co-pay of index medication ≥$50 (vs <$50)	0.92 (0.63–1.34)	0.666
Patients with ≥ 1 hospitalization	0.88 (0.49–1.61)	0.687
Total healthcare pre-index costs	1.16 (0.91–1.48)	0.244

Abbreviations: CI, confidence interval; ICS, inhaled corticosteroids; LABA, long-acting β₂-agonist.

3.4. Persistence

A higher proportion of participants vs nonparticipants were persistent with their index prescription over the 12-month post-index period (12.98% vs 5.73%, p = 0.0044); 19.85% of participants vs 33.59% of nonparticipants (p = 0.0004) discontinued their index medication, and 40.84% of participants vs 29.01% of nonparticipants (p = 0.0045) augmented their index medication (Table 3). The median time to discontinuation or switch from index medication was 365 days for participants and 90 days for nonparticipants (Figure 3). For participants and nonparticipants with at least two prescriptions, the median time for discontinuation or switch from index medication was 365 days and 304 days, respectively (data not shown). Adjusted analyses revealed that participants had a 47% reduction in the risk of discontinuing their index prescription compared with nonparticipants (adjusted hazard ratio [HR]: 0.53; 95% CI: 0.42–0.68; p < 0.0001; Table 5).

Figure 3. KM survival estimates for time (in days) to discontinuation or switch from the index date.

The median time (in days) to discontinuation or switch from the index medication was 365 days for the program participant group and was comparatively lower, i.e. 90 days for the nonparticipant group. The median time was 365 days for the program participant group, as the number of patients who switched or discontinued in this group was < 50%. Following the index date, 14 participants (5% of 262 participants) and 68 nonparticipants (26% of 262 nonparticipants) had only one claim for the index drug. Despite the inclusion of 184 pairs of matched patients, the median time to discontinuation remained significantly higher for participants vs nonparticipants; however, the median time to discontinuation was 304 days for nonparticipants. The drop observed in the nonparticipant group may be due to the higher proportion of nonparticipants switching within 30 days of enrollment.KM, Kaplan-Meier.

Table 5. Risk of discontinuation: Cox proportional hazards model.

Variable	Parameter estimate	Standard error	Chi-square	HR (95% CI)	p value
Program enrollment (key variable)
Program participants (vs nonparticipants)	−0.633	0.124	26.191	0.53 (0.42–0.68)	<0.0001
Comorbidities
Smoking	−0.137	0.126	1.174	0.87 (0.68–1.12)	0.279
Sleep disorders	0.054	0.138	0.154	1.06 (0.81–1.38)	0.695
Elixhauser score
2 (vs 0–1)	0.552	0.197	7.816	1.74 (1.18–2.56)	0.005
3 (vs 0–1)	0.397	0.213	3.458	1.49 (0.98–2.26)	0.063
≥4 (vs 0–1)	0.655	0.209	9.785	1.93 (1.28–2.90)	0.002
ICS/LABA use (pre-index)	0.032	0.123	0.069	1.03 (0.81–1.32)	0.793
Influenza vaccine (pre-index)	−0.291	0.121	5.788	0.75 (0.59–0.95)	0.016
Pneumococcal vaccine (pre-index)	0.048	0.176	0.075	1.05 (0.74–1.48)	0.785
Co-pay of index medication ≥$50 (vs <$50)	−0.075	0.122	0.376	0.93 (0.73–1.18)	0.540
Patients with ≥ 1 hospitalization	0.030	0.199	0.023	1.03 (0.70–1.52)	0.880
Total healthcare pre-index costs	−0.099	0.085	1.351	0.91 (0.77–1.07)	0.245

Abbreviations: CI, confidence interval; HR, hazard ratio; ICS, inhaled corticosteroids; LABA, long-acting β₂-agonist.

3.5. HCRU and costs

3.5.1. All-cause HCRU and costs during the post-index period (postmatching)

The proportion of patients who had utilized health services was similar between participants and nonparticipants for all-cause HCRU over the 12-month post-index period. Over a 12-month period, participants had reduced total mean ($26,515 vs $34,814, p = 0.12) healthcare costs per patient compared with nonparticipants; median costs were $13,995 vs $13,722 (p = 0.77; Figure 4). Supplementary Table S1 shows the mean and median values for all-cause HCRU over the 12-month post-index period. Participants had greater total mean ($11,879 vs $10,946, p = 0.54) and median ($7,836 vs $7,168, p = 0.02) pharmacy expenses. Participants vs nonparticipants had lower total mean ($5,126 vs $11,716, p = 0.12) inpatient costs, respectively. Total mean ($37,306 vs $66,727, p = 0.24) and median ($18,503 vs $37,477, p = 0.03) inpatient costs were lower in patients with ≥ 1 hospitalization (n = 82). While not statistically significant, the proportion of participants with ≥ 1 hospitalization was lower for participants than for nonparticipants (36 [13.74%] vs 46 [17.56%], p = 0.23; data not shown). The overall mean ($9,510 vs $12,152, p = 0.22) and median ($4,043 vs $4,852, p = 0.27) outpatient medical costs per patient were lower for participants vs nonparticipants. Inpatient cost among patients with hospitalization was the key driver of incremental cost disparities between participants and nonparticipants. Adjusted analyses revealed that participants had 24% lower total medical costs (inpatient and outpatient) than nonparticipants, with a cost ratio of 0.76 (95% CI: 0.55–1.04, p = 0.083; Table 6).

Figure 4. Mean ± SD all-cause HCRU and healthcare costs for program participants and nonparticipants (postmatching) assessed over the 12-month post-index period.

Program participants vs nonparticipants showed lower 12-month total mean ($26,515 vs $34,814, p = 0.118412) healthcare costs per patient. Program participants vs nonparticipants had higher total mean pharmacy costs ($11,879 vs $10,946, p = 0.54) and showed lower mean inpatient costs ($5,126 vs $11,716, p = 0.12), respectively. Among inpatient costs among patients with ≥ 1 hospitalization, lower total mean ($37,306 vs $66,727, p = 0.24) was observed for program participants vs nonparticipants, respectively. The total outpatient medical costs per patient showed lower total mean ($9,510 vs $12,152, p = 0.22) for program participants vs nonparticipants, respectively. The primary driver of the incremental cost differences between the program participants and nonparticipants was the inpatient cost.ED, emergency department; HCRU, healthcare resource utilization; hosp, hospitalization; SD, standard deviation; USD, United States dollar.

Table 6. Adjusted mean total medical costs (outpatient + inpatient) during follow-up: GLM with log-link/gamma distribution.

Variable	Parameter estimate	Standard error	Wald 95% CI		Cost ratio	Lower limit	Upper limit	Chi-square	p value
			Lower limit	Upper limit
Program enrollment (key variable)
Program participants vs	−0.28	0.16	−0.60	0.04	0.76	0.55	1.04	3.00	0.083
nonparticipants
Adherence
Adherent (PDC ≥80%) vs nonadherent (PDC <80%)	−0.10	0.18	−0.45	0.24	0.90	0.64	1.27	0.35	0.554
Program enrollment – Adherence interaction	0.23	0.24	−0.24	0.69	1.25	0.79	1.99	0.91	0.340
Comorbidities
Smoking	−0.23	0.12	−0.46	0.01	0.80	0.63	1.01	3.51	0.061
Sleep disorders	−0.20	0.13	−0.46	0.06	0.82	0.63	1.06	2.29	0.130
Elixhauser score
2 (vs 0–1)	−0.22	0.18	−0.57	0.13	0.80	0.57	1.14	1.54	0.214
3 (vs 0–1)	0.06	0.20	−0.32	0.44	1.06	0.73	1.56	0.10	0.754
4+ (vs 0–1)	0.25	0.21	−0.16	0.65	1.28	0.86	1.91	1.43	0.232
ICS/LABA use (pre-index)	0.31	0.12	0.08	0.54	1.36	1.08	1.72	6.75	0.009
Influenza vaccine (pre-index)	0.10	0.11	−0.12	0.33	1.11	0.89	1.39	0.82	0.364
Pneumococcal vaccine (pre-index)	−0.24	0.16	−0.55	0.08	0.79	0.58	1.08	2.18	0.140
Co-pay of index medication ≥$50 (vs <$50)	0.15	0.12	−0.09	0.39	1.16	0.91	1.47	1.49	0.222
Patients with ≥ 1 hospitalization	−0.05	0.18	−0.41	0.30	0.95	0.67	1.36	0.08	0.777
Total healthcare pre-index costs	0.60	0.08	0.45	0.75	1.82	1.56	2.12	58.80	<0.0001

Total healthcare costs capped at 99%.

Abbreviations: CI, confidence interval; GLM, generalized linear model; ICS, inhaled corticosteroids; LABA, long-acting β₂-agonist; PDC, proportion of days covered.

3.5.1.1. COPD-specific HCRU and costs over the post-period (postmatching)

Between participants and nonparticipants, the total healthcare costs per patient showed a higher total mean ($11,376 vs $11,278, p = 0.95) and median ($8,252 vs $6,518, p = 0.0004) in participants (Supplementary Table S2). Similar to the all-cause analysis, the total pharmacy costs were higher for participants vs nonparticipants (mean: $6,696 vs $5,481, p = 0.0003; median $5,774 vs $4,883, p < 0.0001). The total inpatient COPD-specific costs per patient showed higher mean values for participants vs nonparticipants, but the difference was not significant. The total inpatient costs for patients with ≥ 1 hospitalization was lower for participants vs nonparticipants (mean: $24,508 vs $34,306, p = 0.63; median: $11,083 vs $17,487, p = 0.11). Total outpatient medical costs showed a lower mean for participants vs nonparticipants (mean: $3,184 vs $4,468, p = 0.3535; median: $808 vs $781, p = 0.83).

4. Discussion

This study demonstrated that patients who enrolled in the HealthPrize digital behavior change program had significantly increased adherence and persistence compared with matched controls. Studies of digital health interventions are often criticized for failing to control for selection bias, the concern that perhaps only the already motivated or healthier patients enroll. We sought to minimize this concern in our methodology by addressing the ‘healthy adherer’ bias to the best of our ability [20]. Treatment-experienced participants were matched with treatment-experienced nonparticipants given that treatment-naïve patients pose different adherence risks. Well-balanced markers, such as the Elixhauser Comorbidity Index, pharmaceutical claims data, and time from COPD to index date, among others, also helped to address potential bias. And importantly, we controlled for other markers of healthy behaviors available via claims data – screening tests and vaccinations.

Study subjects were significantly more persistent than controls, and it is worth expanding on this adherence marker as it relates to assessing our study in the context of others. Persistence and discontinuation are defined differently by different studies, with the definition dependent upon the allowable number of days late to fill [3,13]. Further, persistence is typically assessed in treatment-naïve populations [3,21] and this study included only treatment-experienced patients. For this reason, persistence rates in the current study may not be appropriately compared with those from other persistence studies of treatment-naïve patients. Nonetheless, even patients on treatment for years are at risk of discontinuing their refills over time, albeit at a much lower rate than treatment-naïve patients [3,21,22]. In one study of treatment-naïve veterans with COPD, a discontinuation rate of 70% was observed within 90 days of prescription [23]. Therefore, adherence interventions are even more crucial in this population.

Medication nonadherence is a complex problem with multiple contributing factors, including unpleasant side effects or the fear of side effects, lack of patient understanding regarding the importance of continuing treatment despite symptomatic improvement, medication cost, forgetfulness, poor provider communication, and gaps in continuity of care [2,6]. Although the RespiPoints™ program did not tailor the patient experience to individual adherence challenges, future interventions could certainly be enhanced by such personalization. That said, single interventions targeting single challenges can be limited in efficacy. For example, one high-profile randomized controlled trial – in cardiovascular disease – aimed to assess the impact of providing free medication to patients discharged from the hospital following acute myocardial infarction. The study showed minimal differences in PDC between control and intervention groups (a 4 to 6 percentage point increase, in comparison to 22 percentage points in the present study) [24]. This challenges the notion that cost is the primary driver of the problem or that single interventions are sufficient.

By contrast, the approach we take to adherence is multifaceted and driven by an understanding of human psychology, recognizing that nonadherence is often driven by limitations in both intrinsic and extrinsic motivation. Even for patients of adequate means and a basic health literacy, many still struggle to act in their own best interest. Similar to other behaviors, such as exercise and saving for retirement, motivation to act in the present for more distant future gains is often challenging, a human foible that behavioral economists call ‘present bias’ [25].

Other studies have demonstrated the link between better adherence and lower healthcare costs [11–15,26]. This study demonstrated numerically lower total costs, but the results did not achieve statistical significance likely due to sampling size constraints. Furthermore, lower healthcare costs were partially offset by slightly higher pharmacy costs possibly attributable to higher adherence.

This study should be interpreted with the following limitations in mind. As with most adherence studies, the data were based on secondary claims data. There could be inaccurate, incomplete, or missing data, such as medication obtained outside of the plan or via physician-administered free samples. Moreover, proof of refill does not guarantee proof of use, although there is no reason to believe that this variable would be different between groups. Further, select variables of importance in real-world datasets [20] were either unmeasured or unavailable to us, resulting in potential residual confounding [27]. For example, claims data do not allow insight into patient income or into other healthy behaviors such as exercise frequency or smoking avoidance, all of which can impact outcomes. As such, potential differences between those who were offered the program but declined compared with those who enrolled were unknown and could not be analyzed.

In addition, this study was limited to COPD patients using TIO or TIO/OLO. Although we cannot confirm that findings would be generalizable to patients using other COPD treatments, it would be logical to assume that the psychological underpinnings of nonadherence are not limited to specific medications or drug classes. Further, based on internal HealthPrize data, similar improvements in adherence have been demonstrated across diverse chronic conditions, indicating a generalizability of efficacy.

Finally, this study used an age-based eligibility criterion of patients ≥18 years. COPD studies often use a 40-year-old age cutoff as an approach to excluding patients with asthma who may have been coded as having COPD. This study explicitly excluded patients with administrative codes related to asthma, and the age distribution showed that there were almost no patients aged <40 years of age in the study.

With prescribed therapies playing such an important role in the management of COPD, and with nonadherence such a major challenge, more attention must be paid to interventions that can improve not only adherence, clinical outcomes, and HCRU but also – even more importantly – quality of life.

5. Conclusion

The HealthPrize behavior change program was associated with a significant improvement in treatment adherence and persistence for adults with COPD. Further evaluation via a randomized controlled trial, including the addition of data related to other healthy behaviors could be even more valuable. While differences in HCRU and cost were encouraging, demonstration of statistical significance was likely compromised by sample size. Further study with larger sample sizes is warranted. Given the complexity of the problem and that traditional adherence interventions often fall short, clinicians, payers, and pharmaceutical companies should consider a broader array of interventions.

6. Expert opinion

Medication nonadherence remains one of the most challenging and pervasive barriers to achieving optimal clinical outcomes. This is particularly true in COPD, with adherence levels generally lower than in other chronic conditions. The 44% increase in adherence achieved by the HealthPrize RespiPoints™ program is higher than results demonstrated by other adherence interventions published to date across a range of chronic conditions. Many traditional interventions, particularly ones targeting only single barriers such as cost or forgetfulness, tend to offer only single-digit benefits.

The success of this more multi-faceted behavior change program, designed with a view through the lens of human psychology, is especially notable given the careful steps taken to match subjects with controls. Research in the digital health world is often criticized for failing to control for ‘selection bias’ – the phenomenon in which the already motivated and conscientious patients are more likely to sign up and engage in various programs. This has the effect of inflating actual efficacy.

Further, in adherence studies, efforts to link an intervention to better outcomes or lower costs are potentially subject to the related ‘healthy adherer’ bias. It can be near impossible to disentangle the effect of medication adherence from that of other critical health-related behaviors. In the present study, subjects and controls were matched not only for pre-index medication use, but also for markers of other adherent behaviors (adherence to screening tests and vaccinations – among the few adherence behaviors discernible via insurance claims alone).

Wider implementation of innovative adherence and other behavior change interventions in healthcare has the tremendous potential to improve outcomes and lower healthcare costs, particularly if aimed at the patients in greatest need. Challenges to such implementation, however, are many. First, there is the ‘who pays?’ problem. Pharmaceutical companies and payors are most likely to support such programs. However, proof of efficacy can be tricky in the real world where large companies tend to implement multiple programs simultaneously, making it difficult for any new single intervention to ‘take credit’ for an improved endpoint. This can lead to promising pilots failing to gain wider commercial traction. In addition, proving the endpoints of better outcomes or lower costs naturally take time, requiring implementation across more than one or two fiscal quarters.

Another challenge is the difficulty of upending the status quo in healthcare. Programs that focus on fun, enjoyment, gamification, or rewards can face an uphill battle, even if they prove more effective than traditional, more ‘boring’ offerings. Sensitivities surrounding the use of financial rewards, in particular, persist despite evidence that even very small rewards can help boost enrollment and engagement (if implemented well). Too many well-meaning, traditional programs have the ‘if you build, it they will come’ assumption, but end up failing to attract patients because there is too little perceived value.

Even small rewards can help overcome enrollment and engagement barriers by adding actual, short-term value from the patient perspective. This is especially important for medication adherence programs, where for many patients a fundamental value problem underlies nonadherence. Consider most chronic medications. A blood pressure pill, for example, offers no short-term value to a patient (but often poses short-term barriers such as a co-pay, a trip to the pharmacy, or transient annoying side effects). The drug’s actual clinical value is long-term, which poses a challenge similar to the difficulties inherent in saving for retirement. Behavioral economists call this challenge ‘present bias.’

And finally, patient selection is key. Patients who demonstrate ‘primary nonadherence’ (those who never even fill a prescription once) should be targeted for intervention along with patients in the ‘secondary nonadherence’ category (already filling and refilling their prescriptions but not taking it often enough). Frustratingly, since only patients in the ‘secondary nonadherence’ category are included in Medicare STAR ratings, they are often the only patients invited into such digital health programs. However, patients who fail to fill at all are naturally at greatest risk of poor outcomes.

Given the plethora of proven and effective pharmaceutical therapies already available, there is a tremendous potential to improve clinical outcomes simply by boosting adherence to existing therapies. And, from a cost effectiveness standpoint, investing in adherence programs is far less expensive than developing new therapeutics. The healthcare industry would be wise to focus on both.

Article highlights

• Medication adherence in COPD was significantly improved by a digital behavior change program.

• Self-selection bias between subjects and controls was minimized by matching for markers of healthy behavior.

• An encouraging trend toward healthcare cost reduction warrants additional study.

• Education, gamification, and insights from behavioral economics, including rewards, can improve medication adherence.

• Although medication adherence in COPD is a significant challenge a novel digital approach can be effective.

Abbreviations

AIDS, acquired immunodeficiency syndrome; CAD, coronary artery disease; CCI, Charlson Comorbidity Index; CE, continuous enrollment; CI, confidence interval; COPD, chronic obstructive pulmonary disease; ED, emergency department; GB, gall bladder; GLM, generalized linear model; GPI-8, Generic Product Identifier-8; HCRU, healthcare resource utilization; HIPAA, Health Insurance Portability and Accountability Act; HIV, human immunodeficiency virus; HR, hazard ratio; ICD, International Classification of Diseases; ICS, inhaled corticosteroids; KM, Kaplan-Meier; LABA, long-acting β₂-agonist; LAMA, long-acting muscarinic antagonist; OLO, olodaterol; PDC, proportion of days covered; PDE-4, phosphodiesterase-4; POS, point-of-service; PPO, preferred provider organization; PSA, prostate-specific antigen; PSM, propensity score matched/matching; SD, standard deviation; SMD, standardized mean difference; TIO, tiotropium bromide; USD, United States dollar

Ethics approval and informed consent

As the data presented in the study are already deidentified, Institutional Review Board (IRB) approval was not required.

Declaration of interest

K S Firlik and V Hayes are employees of HealthPrize Technologies. V Ruthwik Anupindi and M DeKoven are employees of IQVIA and were contracted to conduct this study. A Shaikh was an employee of Boehringer Ingelheim Pharmaceuticals, Inc. at the time of the study and J Franchino-Elder is an employee of Boehringer Ingelheim Pharmaceuticals, Inc. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Author contributions

K S Firlik, V Ruthwik Anupindi, M DeKoven, A Shaikh, and J Franchino-Elder made substantial contributions to the conception and design of the work; V Ruthwik Anupindi, V Hayes, and M DeKoven contributed to the acquisition, analysis, and interpretation of data for the work. All the authors contributed to drafting of the manuscript or revising it critically for important intellectual content and provided final approval of the version submitted for publication. All the authors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The authors meet the criteria for authorship as recommended by the ICMJE. The authors received no direct compensation related to the development of the manuscript.

Abstract and poster presentation details

Abstracts reporting part of these findings have been presented as poster presentations at the Annals of the American Thoracic Society (ATS) 2022 annual meeting on May 13–18, 2022, in San Francisco, U.S.A.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

Writing, editorial support, and formatting assistance were provided by Reham M. Milhem, PhD, of Cactus Life Sciences (part of Cactus Communications), who was contracted and compensated by Boehringer Ingelheim Pharmaceuticals, Inc. (BIPI) for these services. BIPI was given the opportunity to review the manuscript for medical and scientific accuracy as well as intellectual property considerations.

Data availability statement

To ensure independent interpretation of clinical study results and enable authors to fulfill their role and obligations under the International Committee of Medical Journal Editors (ICMJE) criteria, Boehringer Ingelheim grants all external authors access to relevant clinical study data. In adherence with the Boehringer Ingelheim Policy on Transparency and Publication of Clinical Study Data, scientific and medical researchers can request access to clinical study data after the primary manuscript is published in a peer-reviewed journal, regulatory activities are complete, and other criteria are met. Researchers should use the https://vivli.org/ link to request access to study data and visit https://www.mystudywindow.com/msw/datasharing for further information

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/14737167.2023.2296561.

Additional information

Funding

This research was supported by Boehringer Ingelheim Pharmaceuticals, Inc. (BIPI) and HealthPrize Technologies. BIPI was given the opportunity to review the manuscript for medical and scientific accuracy, as it relates to BIPI substances as well as intellectual property considerations.