The economic burden of pleural effusions in patients with chronic myeloid leukemia treated with tyrosine kinase inhibitors

Abstract

Objective:

Tyrosine kinase inhibitors (TKI), the standard of care for patients with chronic myeloid leukemia (CML) patients, may in some cases lead to the development of pleural effusion (PE). The purpose of this study is to compare healthcare resource utilization and costs associated with PE among CML patients treated with a TKI therapy.

Methods:

Two large retrospective claims databases (1999–2009) were combined to identify adult CML patients who received ≥1 TKI prescription before the index date, which was defined as 30 days before the first PE diagnosis for patients with PE and a randomly selected date for PE-free patients. Patients were followed for 6 months after the index date. PE and PE-free patients were matched on a 1:1 ratio. PE-related resource utilization and costs (measured in 2009 US dollars) were estimated for PE patients. All-cause and CML-related resource utilization and costs were compared between PE and PE-free patients. Multivariate regression models were used to control for confounding factors.

Results:

The study included 186 matched pairs. PE-free and PE patients were on average 65.4 and 63.6 years old and 39.8% and 48.9% were female, respectively. PE patients had a significantly higher number of inpatient (IP) days, IP admissions, outpatient (OP) visits and emergency room (ER) visits than PE-free patients (all p < 0.01). All-cause medical services costs were $88,526 and $30,434 for PE and PE-free patients, respectively. After adjusting for confounding factors, the PE-related total medical costs were $47,288 (p < 0.01), which was mostly accounted for by higher IP (difference: $34,123, p < 0.01) and OP (difference: $9563, p < 0.05) costs. PE patients also incurred higher CML-related medical costs compared to PE-free patients (difference: $39,599; p < 0.01).

Conclusion:

PE presents a substantial economic burden for CML patients treated with TKI.

Keywords::

• Chronic myeloid leukemia
• Pleural effusion
• Tyrosine kinase inhibitors
• Medical costs

Introduction

Chronic myeloid leukemia (CML) is a malignancy characterized by the unregulated growth of predominantly myeloid cells and their progenitors in the bone marrow leading to a high count of mature granulocytes and their precursors. It is estimated that ∼24,800 people in the US are currently living with CML, and ∼5150 new cases of CML were expected to be diagnosed in 20111,2.

The use of tyrosine kinase inhibitors (TKIs) that inhibit the enzymatic activity of the BCR-ABL enzyme constitute the mainstay of CML therapy. Three tyrosine kinase inhibitors (TKIs) have been approved by the US Food and Drug Administration (FDA) for the treatment of Philadelphia chromosome-positive (Ph+) CML: imatinib3, nilotinib4, and dasatinib5.

The introduction of imatinib in 2001 dramatically changed the natural history of CML. Imatinib was shown to be more effective than interferon-alpha plus cytarabine (the former standard of care) in preventing disease progression, achieving hematological, cytogenetic, and molecular responses in the phase III multinational IRIS study6,7, and substantially increasing life expectancy compared with historical controls treated with cytarabine/interferon-alpha-based therapy8. CML patients resistant or intolerant to imatinib are generally treated with nilotinib or dasatinib, which are effective, more potent TKIs. Recently, both of these agents have also been approved as first-line therapies for chronic phase CML due to their superior efficacy compared to imatinib in randomized phase III studies4,5.

Although TKIs are revolutionary treatments that have dramatically changed the treatment of CML, they can induce significant side-effects due to off target kinase inhibition; however, not all side-effects are common. Most frequent side-effects of TKIs include peripheral edema, cytopenias, liver toxicity, rash, nausea, vomiting, diarrhea, weight gain, and electrolyte abnormalities. Rarer toxicities include pancreatitis, bleeding, hyperglycemia, and pleural effusion (PE). In patients with chronic phase CML, PE has been infrequently reported with imatinib, with incidence rate ranging from 0–2%, based on clinical trials and an observational study9. Nilotinib is a derivative of imatinib, and it shows a similar incidence rate of PE, occurring in ∼1% of patients in chronic phase CML4,10. However, the occurrence of PE is relatively common with dasatinib, with incidence rates ranging from 14–30% in clinical trials5,11, including ∼4% grade 3–4 PE in chronic phase CML, which is more common in advanced disease10. The rate is particularly high when given in high doses after imatinib failure to patients with advanced phase CML. When used at the currently approved dose schedule of 100 mg once daily, the incidence of pleural effusion decreased markedly compared with the former 70 mg twice daily regimen. In a dose-optimization phase III study, in which the efficacy of dasatinib after imatinib failure was tested in patients with CML in chronic phase, 14% of patients with a minimum of 24-month follow-up receiving dasatinib 100 mg once daily experienced PE (including 2% grade 3/4 PE) compared with 23% in patients treated with dasatinib 50 mg twice daily, 25% for dasatinib 70 mg twice daily, and 26% for dasatinib 140 mg once daily12. Although ∼25% of patients with low-grade PE do not display any symptoms, PE may be associated with symptoms, such as shortness of breath, chest pain, dry cough, or hemoptysis, that can impact patients’ quality-of-life13,14. The management of TKI-induced PE may require the interruption or dose reduction of the TKI and in some cases the use of diuretics and/or steroids9. Grade 3–4 PEs may require invasive procedures such as thoracentesis or pleurodesis to permanently halt the accumulation of fluid in the chest cavity and improve patients’ quality-of-life. However, procedures such as thoracentesis may need to be performed on several occasions15.

According to the National Comprehensive Cancer Network (NCCN) guidelines, it is recommended to interrupt the TKI treatment if a PE occurs. However, treatment interruption in CML patients could potentially impact the efficacy of TKI therapy as well as the patient’s adherence to treatment16–18. A recent study by Marin et al.19 has also shown that poor adherence (≤90%) may impede the achievement of major molecular responses in a great proportion of patients with CML treated with imatinib. Therefore, the occurrence of TKI-associated PE may complicate CML management and could have a substantial economic impact.

The direct costs of TKI-associated PEs have been evaluated in a study estimating resource utilization of 48 dasatinib-treated patients with imatinib resistant/intolerant CML. The study showed that the average cost of treating PE was between $2062 and more than $2700 for all severity levels, and between $6400 and greater than $9000 for those who required invasive procedures (costs in 2007 US dollars)20. However, the method used in that study may have under-estimated the cost of PE because it did not take into account the cost of testing and diagnosing PE, and its impact on the subsequent course of TKI treatment and CML management, which may have higher economic consequences than the direct costs of managing PE. To date, no studies have investigated the overall clinical and economic burden of PE, beyond the direct costs of managing PE itself.

The present study aimed to estimate the overall economic burden of PE in patients with CML treated with any TKI. The study used retrospective administrative claims data to compare overall and CML-related healthcare resource utilization and costs between CML patients who developed PE and matched PE-free CML patients. The findings from this study provide a more comprehensive picture of the economic implications of PE in CML patients treated with a TKI.

Patients and methods

Data sources

This study used retrospective administrative claims data obtained from two sources: the Ingenix/IHCIS Impact National Managed Care database (Ingenix Impact) and the Thomson Reuters MarketScan database. Both are large, nationally representative databases that cover all census regions in the US and contain information on patient demographics, enrollment history, claims for inpatient (IP) and outpatient (OP) medical services, and pharmacy claims. The Ingenix Impact database includes claims submitted by 60 million individuals to more than 46 commercial health plans. The MarketScan database includes ∼25 million individuals annually covered by 130 health plans and self-insured employers. Data are de-identified and comply with the patient confidentiality requirements of the Health Insurance Portability and Accountability Act (HIPAA). Data from January 1999 to June 2009 was extracted from the databases and combined.

Study design

A retrospective matched-cohort design was used to conduct this analysis. Patients were included in the study if they had at least two independent diagnoses for CML (The International Classification of Diseases, 9th Revision, Clinical Modification [ICD-9-CM]: 205.1 x); and had at least one pharmacy claim for a TKI, including imatinib, nilotinib, or dasatinib. Patients were classified into two mutually exclusive study cohorts (PE and PE-free cohorts) based on whether or not they received a diagnosis for PE (ICD-9-CM: 511.9 x) at some point during the entire observation period.

In order to include all the resource utilization and costs associated with diagnosing PE, the index date was defined as 30 days before the first PE diagnosis for patients with PE. For PE-free patients, the index date was randomly selected among all the eligible calendar dates, that is, those following a diagnosis for CML and a prescription for a TKI and followed by at least 6 months of continuous eligibility. Only patients who were at least 18 years of age as of the index date and had continuous healthcare coverage (i.e., eligibility) for at least 6 months before and 6 months after the index date were included.

PE and PE-free patients were matched on a 1:1 ratio using an optimal propensity score matching algorithm. This algorithm included two steps. In the first step, a logistic regression including all eligible PE and PE-free patients was used to generate a propensity score for each patient. The propensity score is the predicted probability of being in the PE group given a set of relevant available variables, which are the covariates of the logistic regression, including age as of the index date, gender, index year, Deyo-adapted Charlson Comorbidity Index (CCI)21, type and total duration of TKI use, CML duration measured prior to the index date, bone marrow or stem cell transplant, diagnosis for CML remission (ICD-9 code 205.11) or relapse (ICD-9 code 205.12), and an exhaustive list of comorbidities and adverse events (AEs) observed during the baseline period (defined as the 6-month period prior to the index date). In the second step, an optimal matching algorithm was employed, where for each PE patient, a single PE-free patient with a similar propensity score was identified, such that the total absolute distance between the PE and the paired PE-free patient was minimized among all possible matched pairs. Patient outcomes were measured during the 6-month period after the index date (Figure 1).

Figure 1.  Study period.

Outcomes measurements and statistical analyses

Patient characteristics

Patient characteristics included demographics (age, gender), index year, CML characteristics (CML duration prior to the index date, diagnoses for CML remission or relapse, and CML complexity measured during the baseline period), CML treatment history (type of TKI use measured both during the baseline period and any time prior to the index date, total TKI use duration, and other CML treatments used prior to the index date), comorbidities (Deyo-adapted CCI and other comorbidities observed during the baseline period), and AEs other than PE.

The duration of CML was calculated as the number of days between the first observed CML diagnosis and the index date. CML disease complexity was classified as mild, moderate, or severe according to a CML complexity algorithm developed by Darkow et al.16 based on diagnoses reported in the medical claims. This measure (clinical construct) reflects diagnoses that would be expected to result in greater difficulty to manage a patient’s disease. The type of CML treatments before the index date included TKIs, bone marrow/stem cell transplant, and other chemotherapies. Total TKI treatment duration was calculated as the number of days between the first prescription for a TKI and the index date. Comorbidities were assessed based on the presence of claims associated with a list of conditions defined by the US Agency for Healthcare Research and Quality22. AEs other than PE were selected from the most common treatment-emergent AEs reported from imatinib, dasatinib, and nilotinib prescribing information23–25. While an exhaustive list of comorbidities were considered, only comorbidities with a prevalence of at least 5% in either cohort were reported. Statistical comparisons between the PE and PE-free patient characteristics were conducted using Wilcoxon signed-rank tests for continuous variables and McNemar tests for categorical variables.

Healthcare resource utilization

The following measures were assessed for each cohort during the study period: number of IP admissions, number of IP days, number of emergency room (ER) visits, number of OP visits, and number of other medical service visits (e.g., laboratory, radiology, or other ancillary services).

All-cause and CML-related (i.e., medical services associated with a CML diagnosis code) resource utilization were reported separately. CML-related resource utilization was defined as medical services associated with a diagnosis for CML. Healthcare resource utilization was compared between PE and PE-free patients using negative binomial regression models. Results were reported as unadjusted and adjusted incidence rate ratios (IRR) with 95% confidence intervals. Adjusted IRRs controlled for potential confounding factors including age, gender, chemotherapy, bone marrow or stem cell transplant, comorbidities with significantly different prevalence at baseline (anemia), CML duration, CCI, any AE at baseline, and CML complexity.

Medical costs

In order to estimate the incremental direct costs associated with PE, all-cause and CML-related healthcare costs incurred by PE and PE-free patients were measured during the 6-month study period and compared between cohorts. Healthcare costs were measured from a managed care perspective and were inflation-adjusted to 2009 US dollars using the consumer price index medical component26. Total healthcare costs were divided into medical costs and pharmacy costs. Medical costs referred to all costs related to IP, OP, ER, and other medical visits. All-cause and CML-related (i.e., medical services associated with a CML diagnosis code) medical costs were reported separately. CML-related OP and ER costs were estimated based on medical services for which CML was either the primary or secondary diagnosis, while CML-related hospitalization costs were calculated by including all medical service costs incurred during a hospitalization for which there was at least one CML diagnosis. Unadjusted all-cause and CML-related cost differences between the PE and PE-free patients were reported and statistical significance was tested using Wilcoxon signed-rank tests. In addition, adjusted all-cause and CML-related cost differences were estimated using multivariate generalized linear models with a gamma distribution and a log link and two-part regression models (for cost components with more than 5% of patients with zero costs) where the first part was a logistic regression model and the second part a generalized linear model with a gamma distribution and a log link. p-values were estimated using a non-parametric bootstrap resampling technique of 500 iterations. All multivariate regression models controlled for age, gender, chemotherapy, bone marrow or stem cell transplant, comorbidities with significantly different prevalence at baseline (anemia), CML duration, CCI, any AE at baseline, and CML complexity.

General statistical considerations

Statistical significance was set at a two-sided alpha-level of 0.05 or less. All statistical analyses were performed using SAS version 9.2 (SAS Institute, Inc., Cary, NC).

Results

Population characteristics

A total of 186 CML patients in each cohort were matched (Figure 2). On average, PE-free and PE patients were 65.4 and 63.6 years old, and 39.8% and 48.9% were female, respectively (Table 1). After matching, both cohorts were found to be balanced in terms of demographic characteristics, CML duration, CML complexity, chemotherapy and TKI use (Table 1), their comorbidity profile, and prevalence of AEs at baseline (Table 2). However, despite patient matching, the proportion of patients who received other chemotherapy at baseline, who received bone marrow or stem cell transplant during the baseline period, and who experienced a hematologic adverse event during the baseline period remained significantly higher in the PE cohort (Tables 1 and 2). These differences were adjusted for in the multivariate regression models.

Figure 2.  Patient selection flowchart.

Table 1.  Matched patient baseline characteristics.

Matching criteria	PE-free patients (n = 186)	PE patients (n = 186)	p-value
Demographics
Age (years; mean ± SD)	65.4 ± 13.9	63.6 ± 15.0	0.3988
Female, n (%)	74 (39.8%)	91 (48.9%)	0.0760
Index year, n (%)
2005	37 (19.9%)	37 (19.9%)	–
2006	27 (14.5%)	37 (19.9%)	0.1695
2007	36 (19.4%)	44 (23.7%)	0.3127
2008	62 (33.3%)	42 (22.6%)	0.0209*
CML characteristics
CML duration (days; mean ± SD)	586.9 ± 462.2	608.8 ± 466.6	0.5087
CML remission^†, n (%)	33 (17.7%)	48 (25.8%)	0.0595
CML relapse, n (%)	1 (0.5%)	0 (0.0%)	–
CML complexity, n (%)
Mild	123 (66.1%)	124 (66.7%)	0.9126
Moderate	48 (25.8%)	45 (24.2%)	0.7194
Severe	15 (8.1%)	17 (9.1%)	0.7115
CML treatment characteristics
TKI use any tme prior to the index date, n (%)
Dasatinib	20 (10.8%)	24 (12.9%)	0.5207
Imatinib	184 (98.9%)	182 (97.8%)	0.4104
Nilotinib	1 (0.5%)	5 (2.7%)	0.0997
TKI use during the baseline period, n (%)
Dasatinib	20 (10.8%)	19 (10.2%)	0.8656
Imatinib	137 (73.7%)	158 (84.9%)	0.0072
Nilotinib	5 (2.7%)	1 (0.5%)	0.0997
TKI treatment duration, (days; M ± SD)	570.5 ± 446.4	585.2 ± 453.2	0.7010
Other chemotherapy use, n (%)	26 (14.0%)	43 (23.1%)	0.0234*
Bone marrow or stem cell transplant, n (%)	10 (5.4%)	27 (14.5%)	0.0032*

* Significant at the 5% level.

† Based on ICD-9 code 205.x1.

Table 2.  Baseline comorbidities and adverse events.

Matching criteria	PE-free patients (n = 186)	PE patients (n = 186)	p-value
Comorbidity profile
Deyo-adapted CCI (mean ± SD)	2.81 ± 1.35	2.83 ± 1.41	0.9462
General comorbidities, n (%)
Anemia	76 (40.9%)	95 (51.1%)	0.0481*
Cardiovascular diseases	33 (17.7%)	43 (23.1%)	0.1985
Cerebrovascular diseases	14 (7.5%)	11 (5.9%)	0.5344
Chronic pulmonary disease	24 (12.9%)	27 (14.5%)	0.6511
Coagulopathy	10 (5.4%)	20 (10.8%)	0.0569
Depression	7 (3.8%)	12 (6.5%)	0.2390
Diabetes	36 (19.4%)	39 (21.0%)	0.6982
Fluid electrolyte disorders	13 (7.0%)	18 (9.7%)	0.3483
Hyperlipidemia	29 (15.6%)	31 (16.7%)	0.7780
Hypertension	47 (25.3%)	64 (34.4%)	0.0541
Lymphoma	9 (4.8%)	16 (8.6%)	0.1472
Macular degeneration	4 (2.2%)	11 (5.9%)	0.0650
Peripheral vascular disease	15 (8.1%)	12 (6.5%)	0.5488
Renal failure	8 (4.3%)	12 (6.5%)	0.3578
Solid tumor	17 (9.1%)	27 (14.5%)	0.1084
Valvular disease	12 (6.5%)	19 (10.2%)	0.1891
Adverse events, n (%)	125 (67.2%)	133 (71.5%)	0.3683
Fluid retention	12 (6.5%)	20 (10.8%)	0.1391
Hematologic adverse events	60 (32.3%)	79 (42.5%)	0.0417*
Other non-hematologic adverse events	103 (55.4%)	109 (58.6%)	0.5298

* Significant at the 5% level.

All-cause and CML-related resource utilization and costs

For all-cause healthcare utilization, PE patients had more than 6-times as many IP days (IRR = 6.20; p < 0.0001), 2.54-times as many IP admissions (IRR = 2.54; p < 0.0001), and almost twice as many OP visits (IRR = 1.72; p < 0.0001) and ER visits (IRR = 1.87; p < 0.005) compared to PE-free patients (Table 3). For CML-related healthcare utilization, PE patients had almost 7-times as many IP days (IRR = 6.98; p = 0.0002), more than 3-times as many IP admissions (IRR = 3.34; p < 0.0001), and twice as many OP visits (IRR = 2.02; p < 0.0001) compared to PE-free patients (Table 3).

Table 3.  Comparison of resource utilization between PE and PE-free CML patients.

Medical events	Incidence rate (per patient-year)		Unadjusted		Adjusted
	PE patients	PE-free patients	IRR^† (95% CIs)	p-value	IRR^† (95% CIs)	p-value
All-cause related healthcare resource utilization
IP admissions	1.65	0.53	3.10 (2.35–4.10)	<0.0001*	2.54 (1.93–3.34)	<0.0001*
IP days	17.37	3.20	5.42 (3.10–9.47)	<0.0001*	6.20 (3.04–12.67)	<0.0001*
ER visits	0.44	0.17	2.56 (1.57–4.18)	0.0002*	1.87 (1.26–2.77)	0.0018*
OP visits	20.42	13.10	1.56 (1.35–1.80)	<0.0001*	1.72 (1.45–2.06)	<0.0001*
Other medical services	5.73	1.83	3.14 (2.28–4.30)	<0.0001*	2.96 (2.13–4.13)	<0.0001*
CML-related healthcare resource utilization
IP admissions	1.01	0.20	4.95 (3.01–8.12)	<0.0001*	3.34 (2.18–5.11)	<0.0001*
IP days	12.11	2.18	5.56 (2.60–11.89)	<0.0001*	6.98 (2.48–19.62)	0.0002*
ER visits	0.04	0.01	3.50 (0.66–18.56)	0.1411	1.19 (0.34–4.14)	0.7895
OP visits	8.08	4.11	1.96 (1.53–2.51)	<0.0001*	2.02 (1.51–2.71)	<0.0001*
Other nedical services	1.45	0.33	4.41 (2.48–7.85)	<0.0001*	3.57 (2.04–6.25)	<0.0001*

* Significant at the 5% level.

^† An IRR >1 indicates that PE patients had higher incidence of incurring medical services compared to PE-free patients.

Within the 6 months of the study period, PE patients incurred, on average, total healthcare costs of $105,715, of which $55,080 (52.1%) were CML-related medical service costs and $17,189 (16.3%) were pharmacy costs ($12,800 or 12.1% for TKIs and $4389 or 4.2% for other drugs). PE-free CML patients incurred total healthcare costs of $49,621 on average, of which $15,536 (31.3%) were CML-related medical service costs, and $19,186 (38.7%) were pharmacy costs ($17,357 or 35.0% for TKIs and $1830 or 3.7% for other drugs). Compared with PE-free patients, PE patients incurred total medical service costs $47,288 higher over a 6-month period (adjusted costs; p < 0.0001; Table 4), which was mostly due to PE patients’ higher IP and OP costs ($34,123, p < 0.0001 and $9563, p < 0.05; respectively, Table 4). Similarly, PE patients incurred higher CML-related medical costs compared with PE-free patients (difference: $39,599, p < 0.0001; Table 4), due to CML-related higher IP and OP costs ($21,557, p < 0.0001 and $9146, p < 0.05, respectively; Table 4). PE patients incurred lower TKI costs compared to PE-free patients (Table 4), potentially because treatment interruption is recommended in the case of PE (adjusted difference: $3692; p < 0.0001). On the other hand, PE patients incurred significantly higher costs for other drugs compared to PE-free patients (adjusted difference: $1913; p < 0.0001; Table 4).

Table 4.  Comparison of healthcare costs between PE and PE-free patients.

Medical costs	Healthcare costs		Unadjusted		Adjusted
	PE patients (mean ± SD)	PE-free patients (mean ± SD)	Cost difference^†	p-value	Cost difference^†	p-value
All-cause related costs
Medical services	88,526 ± 133,474	30,434 ± 61,166	58,092	< 0.0001*	47,288	<0.0001*
IP	48,884 ± 83,525	7,901 ± 29,399	40,984	<0.0001*	34,123	<0.0001*
ER	721 ± 2,168	439 ± 2,333	282	<0.0001*	391	0.172
OP	34,916 ± 73,214	20,156 ± 37,917	14,760	0.002*	9,563	0.016*
Other medical services	4,005 ± 10,384	1,939 ± 7,468	2,066	<0.0001*	1,986	0.012*
Pharmacy	17,189 ± 12,466	19,186 ± 14,866	−1,998	0.389	−1,326	0.391
TKIs	12,800 ± 11,499	17,357 ± 14,589	−4,557	0.003*	−3,692	<0.0001*
Other drugs	4,389 ± 6,604	1,830 ± 1,925	2,559	<0.0001*	1,913	<0.0001*
Total (Medical + Pharmacy)	105,715 ± 133,982	49,621 ± 62,744	56,094	<0.0001*	43,668	<0.0001*
CML-related costs
Medical services	55,080 ± 106,423	15,536 ± 49,714	39,544	<0.0001*	39,599	<0.0001*
IP	34,461 ± 73,706	5,741 ± 28,184	28,720	<0.0001*	21,557	<0.0001*
ER	71 ± 515	30 ± 291	41	0.256	−7	0.918
OP	19,647 ± 59,397	9,316 ± 26,189	10,331	0.029*	9,146	0.024*
Other medical services	901 ± 4,761	448 ± 2,549	453	5 E-04*	505	0.116
Total (Medical + Pharmacy^‡)	67,879 ± 104,457	32,892 ± 50,031	34,987	<0.0001*	26,404	<0.0001*

*Significant at the 5% level.

^†Cost differences >0 indicate that PE patients incurred higher costs compared to PE-free patients.

^‡CML pharmacy costs include TKI costs.

Discussion

The study findings suggest that TKI-associated PE represents a significant economic burden to payers in disease management of CML patients treated with TKIs. Total medical costs were much higher for PE patients compared to PE-free patients ($88,526 vs $30,434); this difference was due in large part to the higher cost of CML-related medical services (IP days, OP visits, and ER visits) for PE patients because they had a higher rate of utilization of these medical services. Interestingly, the majority (62%) of all medical service costs incurred by PE patients was CML-related.

Previous studies have found that management of PE in CML patients is costly, requires intensive resource utilization, and may become an important factor in treatment selection. For example, Stephens et al.20 found that invasive procedures and PE recurrence were the key cost drivers in the management of PE on 48 patients from a large US cancer center. They found that 60% of PEs were mild and moderate and could be managed medically (cost: $750 per episode; in 2007 US dollars), while 40% of PEs were more severe, with half of those requiring invasive procedures (cost: $6400 to >$9000 per episode, in patients without or with a pleural catheter, respectively). In that population, the average per-patient cost of treating PE was $2062–$2700 for all severity levels, which amounts to $2197–$2876 when adjusted to 2009 US dollars using the CPI medical component. These estimates represent only 8.3–10.9% of the cost difference between PE and PE-free patients found in our study. In contrast to Stephens et al., where only PE-related procedure costs were included, our study included costs beyond the simple management of PE, such as the costs of monitoring, testing, diagnosing, and treating, and the costs associated with follow-up management of CML patients after experiencing PE.

According to the NCCN guidelines, TKI treatment interruption is recommended in TKI-treated patients experiencing PE. In one clinical trial, 83% of dasatinib-treated patients with PE required a temporary interruption of dasatinib for a median of 27 days, 71% required a dose reduction, and 6% had permanently discontinued from dasatinib due to recurrent PEs27. Further research using real-world data can further confirm whether occurrence of PE affects the adherence to TKIs in general. TKI non-adherence exposes CML patients at risk for sub-optimal response or disease progression, complicates the management of CML28, and is associated with increased healthcare costs and additional patient morbidity17. One study showed that a 10% point decrease in treatment adherence results in a 14% increase in healthcare costs15. Additionally, patients with adherence lower than 50% had healthcare costs ∼3-times higher compared to patients with adherence to their treatment greater than 95%15. In the current study, patients with PE incurred lower TKI costs compared to PE-free patients, indicating reduced use of TKIs. Further studies are needed to evaluate how PE events impact on treatment adherence and therapy disruption and changes in a real-world setting. The findings of this study suggest that the costs of treating PE may only represent a fraction of the costs consequent to the PE event, and that the cost associated with a more complicated CML management in patients who experienced PE represents the major portion of the incremental healthcare costs compared to patients without PE.

Although the side-effects of PE are typically reversible and manageable, they complicate CML disease management, increase resource utilization and overall healthcare costs, and increase the economic burden to payers. Although the clinical impact of PE on disease progression and overall patient survival are unknown, it has been shown that PE decreases CML patients’ quality-of-life15,16,29. These findings point to the necessity of considering co-morbid conditions that may increase the risk of developing PE, such as poorly controlled hypertension or congestive heart failure26,30,31.

In addition to the estimated PE costs using real world data, it is also important to know the incidence rate of PE in a real-world setting. Although PE occurrence rates have been reported in clinical trials, rates in the real world may be different. Future research using chart review data can further address this question, especially in a comparative study among different TKIs. In addition, predictors of PE can also be studied using a survival analysis framework.

This study is subject to some limitations. PE and PE-free patients may have different disease characteristics and pre-disposing risk factors, potentially confounding the outcomes comparison between the cohorts. To reduce this potential bias, PE and PE-free patients were matched using the optimal propensity score matching method, controlling for a large number of observable clinical variables available in this claims data. In addition, multivariate regression analyses were conducted to control for any remaining differences of baseline characteristics after the matching process. Nevertheless, an unobserved confounding effect may still exist such that the incremental costs of the PE patients may be due to an unobserved worse disease profile, despite the use of both propensity score matching and regression methods, including a large number of observable clinical variables and patient characteristics. Specifically, information on CML phase (i.e., chronic phase, accelerated phase, or blast phase) was not available in the claims databases. The complexity of CML that has been estimated using an algorithm developed by Darkow et al.16 (mild, moderate, severe CML complexity) can be considered as a surrogate for CML severity. Despite the effort of balancing the patient severity at baseline, it is possible that a larger proportion of patients with PE have been in accelerated or blast phase. Among the TKIs, only dasatinib is approved by the FDA for the treatment of patients with blast phase CML, which has been associated with a higher rate of PE compared with chronic phase CML. In addition, PE events are more likely to occur in patients with advanced disease26,32 and, therefore, it is possible that more patients in the PE cohort may have advanced CML, a sub-set of patients whose management is associated with a higher cost, regardless of the presence of PE. In this study, patients in the two cohorts were matched based on their propensity score in addition to the regression models that were applied to further control for residual differences between cohorts, where both methods controlled for a large number of patient characteristics and clinical information observable in the claims database. However, there may still be differences in the clinical profile between the two study cohorts. For example, although index year was controlled for as a covariate, the long retrospective observation period may be associated with changes in clinical practice that cannot be fully balanced based on index year. The study included AEs in the propensity score generation process. However, the AEs may have reporting bias in hospitalized vs non-hospitalized patients that are not further examined in this study. Despite the effect of propensity score matching and regression control, the direction that the PE cohort is associated with more comorbidities and adverse events at baseline than the non-PE cohort implied the non-observable clinical difference would also be more severe in the PE cohort. Future studies based on alternative data sources are warranted to confirm the findings of this study.

Another limitation of this study is that, given that the objective of the present study is to focus on the incremental healthcare utilization and costs, other PE-related treatment and outcomes information such as the types of treatments for PE and treatment pattern changes following a PE event (e.g., TKI dose reduction or interruption and drug switching) were not studied. A further study on the treatment pattern, and short-term and long-term clinical outcomes of PE and its impact on CML disease course may be necessary to provide additional insight on the reasons for the increased healthcare resource use and costs associated with PE. Furthermore, due to the limitation of the claims database that cannot differentiate PE events of different severity levels, the study estimated the costs of recorded PEs on average only, without stratifying by severity. Further research using an alternative database where PE severity information and costs are available can be of interest to further strengthen the findings of this study.

Conclusion

PE is associated with high resource utilization and total healthcare costs among CML patients treated with TKIs.

Transparency

Declaration of funding

This research was sponsored by the Novartis Pharmaceuticals Corporation, which manufactures nilotnib and imatinib.

Declaration of financial/other relationships

DW and AG are current employees of the Novartis Pharmaceuticals Corporation, VB worked for the Novartis Pharmaceuticals Corporation while performing this analysis but is currently employed by Sunovion Pharmaceuticals Inc. Sunovion Pharmaceuticals Inc. was not in any way associated with this study. AG and EW are current employees of Analysis Group, Inc., which has received consultancy fees from the Novartis Pharmaceuticals Corporation. DPL worked for Analysis Group, Inc. while performing this analysis. AQ-C is an employee of the MD Anderson Cancer Center and has received consultancy fees from Novartis Pharmaceuticals Corporation. He also has consultant or advisory relationship with Bristal Myers Squibb. The peer reviewers on this manuscript have disclosed that they have no relevant financial relationships.

Acknowledgments

No assistance in the preparation of this article is to be declared.