Clinical Trajectories, Healthcare Resource Use, and Costs of Long-term Hematopoietic Stem Cell Transplantation Survivors: A Latent Class Analysis

Abstract

Purpose:

To identify patterns of healthcare utilization in allogeneic and autologous hematopoietic stem cell transplantation (HSCT) recipients and evaluate factors associated with high-need and high-cost post-transplantation care.

Methods:

Latent class analysis of a retrospective cohort of long-term allogeneic (n=436) and autologous (n=888) HSCT survivors within the Truven MarketScan database (2009–2014). We assessed factors associated with the latent classes by comparing post-transplantation healthcare utilization including inpatient admissions and length of stay, emergency room visits, specialist visits and primary care provider visits.

Results:

Four utilization classes were identified in allogeneic and autologous HSCT recipients: (i) outpatient specialist care dominant (51.8% and 57.3%); (ii) outpatient primary care dominant (10.3% and 25.7%); (iii) outpatient/inpatient balanced (20.6% and 13.5%); and (iv) inpatient dominant (17.2% and 3.5%). Mean monthly healthcare expenditures in the inpatient dominant utilization class were $41,097 and $25,556 for allogeneic and autologous survivors, respectively, which were two to five times higher compared to other classes during the two-year post-transplantation period. Factors associated with the high utilization class were transfusion (OR=1.87, 95% CI 1.06–3.30) and 100-day post-transplant graft-versus-host-disease (OR=1.76, 95% CI 1.05–2.94) in allogeneic HSCT; higher baseline Charlson comorbidity index (OR=1.45, 95% CI 1.19–1.76) in autologous HSCT.

Conclusion:

Based on distinct patterns of healthcare utilization following HSCT, we identified factors associated with higher resource utilization and greater healthcare related expenditures.

Implications for Cancer Survivors:

Earlier identification of high-cost and high-need HSCT long-term survivors could pave the way for clinicians to offer more continuous engagement in survivorship care delivery.

Introduction

A key challenge for contemporary hematopoietic stem cell transplantation (HSCT) is how to identify subgroups of patients with distinct healthcare needs and to provide individualized post-transplantation care. Patients undergoing HSCT for treatment of hematological malignancies receive health care services from multiple providers and at various levels of intensity in the years following transplantation [1]. Given the nature of these diseases and treatments, patients often require inpatient care when disease relapses or in the event of severe post-transplantation complications. On the other hand, patients achieving satisfactory chimerism without complications may exhibit a pattern of decreasing post-transplantation healthcare resource utilization, eventually transitioning back to primary care providers and pre-transplantation interaction levels with the healthcare system.

While HSCT patients in clinical trials receive high levels of care and follow up, evidence from population-based studies remains limited on the level of health services required to manage HSCT recipients beyond the 100-day period post-stem cell infusion [2, 3]. Knowledge on patterns of healthcare utilization is critical in planning and designing post-transplantation care. Furthermore, because multidisciplinary care coordination among multiple types of care providers and settings are often necessary in managing long-term survivorship, understanding factors related to varying patterns and types of services used could help optimize allocation of healthcare resources and care coordination.

The limited literature on health service utilization has primarily focused on the experience of patients with breast cancer, mental health or chronic disease conditions [4–7]. Overall, higher disease burden prior to cancer diagnosis, knowledge of available services, higher financial capacity and means, and availability of transportation were associated with higher likelihood of post-treatment health services use [8, 9]. Several studies performed using administrative health claims databases and institutional cohorts have investigated the economic aspects of HSCT. Majhail et al described that between 2007 and 2009, the median 100-day total cost for first allogeneic HSCT among privately insured patients was $203,026 (interquartile range [IQR] $141,742-$316,426); for autologous HCT, the median cost was $99,899 (IQR $73,914-$140,555) [10]. A recent study of 1,562 patients undergoing HSCT estimated the 100-day cost to be $289,283 for myeloablative allogeneic HSCT and $253,467 for nonmyeloablative, reduced-intensity conditioning (RIC) allogeneic HSCT compared with $140,792 for myeloablative autologous HSCT [11]. Similarly, the comparison of 100-day overall median costs between two institutions for these regimens were $129,000 (IQR: $84,000–$171,000) and $96,000 (IQR $74,000–$152,000) in 2010 dollars respectively. The 2-year median costs in long-term survivors after RIC HSCT were $39,000 (IQR $19,000–$101,000) [12]. Many studies have described the experience of single institutions treating varied populations with high levels of heterogeneity in treatment regimens and indications for HSCT [13]. In addition, most researchers used median or mean costs to describe the skewed cost distributions for HSCT and these adopted methods were unable to identify subgroups of HSCT recipients requiring consistently high healthcare utilization. Also, many single institutional analyses suffer from small sample sizes and noncontemporary data.

The current evidence on this topic focuses primarily on factors associated with single types of service use (e.g., hospital readmission, emergency room (ER), primary care provider (PCP) visits). Few studies have considered these as a combination of multiple types of healthcare services. It is conceivable that different types of services could be complementary or substitutive. For example, frequent ER visits may be indicative of poor acute disease management and lack of access to specialists or PCPs in outpatient settings. ER visits could also lead to inpatient admission if patients require additional substantial care, and subsequent long-term hospitalization would effectively preclude management of their disease in outpatient settings. Thus, additional research with data-driven approaches characterizing distinct patterns of service use incorporating a wide range of service types among patients with hematological malignancies undergoing HSCT is warranted.

We aimed to identify subgroups of HSCT long-term survivors who share similar patterns of service use after a 100-day post-transplant period using latent class analysis (LCA). We also examined demographic and clinical factors related to post-transplantation latent classification, resource utilization during the first two years following HSCT and the corresponding health care costs incurred across latent classes.

Methods

Design and Data Source

We performed a population-based retrospective cohort study using the Truven MarketScan Health Databases [14]. The database includes a nationally representative sample with information on inpatient, outpatient and pharmacy dispensing records from more than 100 million commercially insured patients. The data are de-identified per our institutional data use agreement and the institutional review board of the University of Illinois at Chicago determined this study to be exempt from human subjects research requiring informed consent.

Study Population

Our analyses initially included a total of 774,113 patients with at least one claim containing an International Classification of Disease-9 (ICD-9) diagnosis or procedure code indicating hematological malignancies (acute and chronic leukemia, lymphoma, myeloma, myelodysplastic syndrome (MDS) and myeloproliferative neoplasms) from January 2009 through December 2014. We included adult patients (18–64 years old) with at least one inpatient diagnosis or two outpatient diagnoses that were at least 30 days apart indicating hematological malignancies. Eligible patients all underwent allogeneic HSCT (ICD-9 procedure code, 41.02, 41.03, 41.05, 41.06, 41.08; Current Procedural Terminology (CPT), 38240) or autologous HSCT (ICD-9 procedure code, 41.01, 41.04, 41.07, 41.09; CPT, 38241) after cancer diagnosis. The date of stem cell infusion was defined as the index date. All patients had at least 180 and 730-day continuous enrollment before and after the index date, respectively.

Patients with more than one HSCT event or more than one type of hematological malignancy diagnoses were excluded from our analysis.

Measures

We measured five different types of healthcare services use to determine services usage patterns: inpatient admission events, length of inpatient stay, ER visits, primary care visits and specialist care visits. All measures of healthcare services use were collected over the 101–365 days and 366–730 days post-transplant period. Measured length of stay was truncated if the hospitalization overlapped with the 365- and 730-day cut-off point. Primary care visits were defined as outpatient services provided by medical doctors, internal medicine physicians, family medicine doctors, nurse practitioners and physician assistants. Specialty care visits were indicated by service provider of hematologist and oncologist. Dates of healthcare service use were identified through claims and same types of services occurring on the same day were counted only once. Acute care utilization rates for inpatient admissions and ER visits were characterized as groups based on number of encounters per period (0, 1, and 2+). We also report mean monthly length of stay (in days), specialist care visits and primary care visits as continuous variables which we further categorized into groups of 0, 1, and more than 1 per month.

The costs of inpatient and outpatient medical encounters and prescription medications were adjusted for inflation into 2014 U.S. dollars, using the medical care component of the U.S. Department of Labor Bureau of Labor Statistics Consumer Price Index (CPI).[15] The total costs incurred over the 101–365 days and 366–730 days periods were then aggregated.

Demographic and clinical variables included age, sex, region, type of hematological malignancies, and baseline chronic health conditions including diabetes mellitus, hypertension and dyslipidemia. These three conditions were included because patients demonstrated high rates of nonadherence and discontinuation of their medication over the follow-up period. The resulting complications could influence the healthcare utilization patterns among HSCT recipients [16]. The National Cancer Institute Charlson comorbidity index (NCI-CCI) was calculated over the 6-month pre-transplantation baseline period [17, 18]. Information was collected on transplant-related variables including transfusion, mucositis, opportunistic infection including cytomegalovirus infection (CMV), candidiasis, aspergillosis, cryptococcosis, and pneumocystis jiroveci pneumonia. Occurrence of graft-versus-host disease (GVHD) was measured for allogeneic HSCT recipients over the 100-day post-transplant period.

Statistical Analysis

Descriptive and inferential statistics were used to compare baseline demographic and clinical characteristics, healthcare utilization, and healthcare expenditures across latent classes within allogeneic and autologous HSCT cohorts. Continuous variables were summarized as means and standard deviations. Categorical variables were summarized as counts and percentages. Statistical comparisons between groups were performed using one-way analysis of variance for continuous variables and χ² test for categorical variables.

We used latent class analysis models to estimate the patterns of healthcare services use in this sample of long-term HSCT survivors [19, 20]. Five types of healthcare service use variables over the 101–365 days period were used as latent class indicators. Model fit indices such as Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and adjusted BIC were calculated to determine the best-fitting sub-class structure, with smaller values indicating better fit. Entropy indices were used for assessing the precision of assigning latent class membership, ranging from 0 to 1, with a value approaching 1.0 being more desirable [19]. A series of models with increasing number of classes ranging from 1 to 6 were assessed to determine the optimal number of latent classes using model fit indices, parsimony considerations, theoretical justification and interpretability [21]. Selection of the final LCA model was based upon these measures varied within a plausible range and clinical judgement.

Once the optimal number of latent classes was defined, we used multivariable logistic regression models to estimate odds ratios (OR) and 95% confidence intervals (CI) for associations between baseline demographic, clinical characteristics and transplant-related covariates with membership in the high utilization latent class. We first assessed associations between predictors and class memberships using univariate analyses and then in fully adjusted multivariable analyses.

We evaluated the correlation between post-transplantation class membership on healthcare expenditures by service types and health service use through the second year post-transplantation. Statistical differences were compared using one-way analysis of variance.

All statistical tests were two-tailed and a p value of <0.05 was defined a priori as statistically significant. Analyses were conducted using SAS version 9.4 (SAS Institute, Inc., Cary, NC) and PROC LCA, version 1.3.2 in SAS [22].

Results

A cohort of consecutive long-term allogeneic HSCT survivors (n=436) and autologous HSCT survivors (n=888) between 2009 and 2014 were identified. (Supplemental Figure 1) Descriptive characteristics of HSCT patients are reported in Supplemental Table 1. The mean age of the allogeneic and autologous HSCT recipients was 49.7 (±11.8) and 52.7 (±9.7) years, respectively. The proportion of male patients were slightly higher in both the allogeneic (52%) and autologous (58%) HSCT groups.

A majority of allogeneic HSCT recipients (73.4%) had diagnoses of acute leukemia or MDS, while most of the autologous HSCT recipients (93.3%) had diagnoses of multiple myeloma or lymphoma. The pre-transplantation NCI-CCI was 3 and above in 178 (40.8%) patients in the allogeneic HSCT group, whereas 404 (45.5%) patients in the autologous HSCT group had scores of 3 and above. The prevalence rates of diabetes, hypertension and dyslipidemia were 15.6%, 31.0%, and 23.2% in the allogeneic HSCT cohort and 14.5%, 41.8% and 23.3% in the autologous HSCT cohort.

Interpretation of Latent Classes and LCA models

Fit indices (AIC and BIC), entropy are generated from pre-specified LCA models with varying number of classes. The value of AIC and BIC reached a minimum in the three and four-class models. Based on the elbow-sharped curves, we concluded that there was no substantial improvement in model fit beyond four-class models as well as a clinically meaningful interpretation with acceptable fit for both allogeneic and autologous HSCT cohorts. In addition, the four-class model has high entropy indicating satisfactory individual-level class membership accuracy [23]. Thus, we selected the four-class models for our main analysis.

Four distinct patterns of healthcare services use were identified with LCA models using conditional probability distribution of indicator variables across subgroups (Figure 1). An estimated 51.8% and 57.3% of allogeneic and autologous HSCT recipients had low acute care utilization and high use of specialist outpatient care. We labeled this group as having “outpatient specialist care dominant pattern”. About 10.3% and 25.7% of the allogeneic and autologous HSCT recipients had low utilization of acute care and high use of primary care, we labeled this group as “outpatient primary care dominant pattern”. The third group had high use of both acute care and outpatient health service, and this group was termed as “outpatient/inpatient balanced pattern”, which accounted for 20.6% and 13.5% of the allogeneic and autologous HSCT recipients. We labeled the remaining 17.2% and 3.5% patients from allogeneic and autologous HSCT cohorts as “inpatient dominant pattern” because their health use pattern was characterized with high acute care utilization including hospital admissions and ER visits.

Figure 1:

Conditional probability of health services use indicator by 4 latent classes. A, The estimated conditional probability (Y-axis) for each indicator variable (X-axis) in the latent class analysis for allogeneic HSCT long-term survivors. B, Presents the same latent classes of the autologous HSCT survivors

OPT, outpatient; IPT, inpatient; ER, emergency room; PCP, primary care provider.

Types of hematological malignancies, baseline and 0–100 day post-transplant period comorbidities are reported in Table 1. The overall incidence of graft-versus-host disease was 38.8% in the allogeneic HSCT group during the 100-day post-transplant period. A higher percentage of allogeneic HSCT recipients acquired opportunistic infection than the autologous HSCT recipients (21.3% vs. 4.8%). Greater proportion of allogeneic HSCT recipients received transfusion compared to autologous HSCT recipients (20.2% vs. 10.1%). No statistically significant differences were observed across subgroups over the 100-day post-transplant period. Differences in health service utilization and healthcare related expenditures by category across these four patterns are presented in Table 2.

Table 1:

Baseline characteristics for the allogeneic HSCT and autologous HSCT long-term survivors according to health service utilization patterns

	Allogeneic HSCT recipients(n = 436)								Autologous HSCT recipients(n = 888)
	Class 1 OPT Specialist (n = 226, 51.8%)		Class 2 OPT PCP (n = 45, 10.3%)		Class 3 OPT/IPT balanced (n = 90, 20.6%)		Class 4 IPT dominant (n = 75, 17.2%)		Class 1 OPT Specialist (n = 572, 57.3%)		Class 2 OPT PCP (n = 165, 25.7%)		Class 3 OPT/IPT balanced (n = 118, 13.5%)		Class 4 IPT dominant (n = 33, 3.5%)
	N	%	N	%	N	%	N	%	N	%	N	%	N	%	N	%
Mean (SD)	49.9	11.5	51.2	11.7	48.5	12.3	50.0	12.4	52.23	9.63	54.42	8.87	52.73	10.54	52.88	10.52
Median	53		54		52		55		55		57		55.5		55
Gender
Male	122	53.98	20	44.44	45	50.00	38	50.67	333	58.22	104	63.03	66	55.93	15	45.45
Female	104	46.02	25	55.56	45	50.00	37	49.33	239	41.78	61	36.97	52	44.07	18	54.55
Region
Northeast	50	22.12	8	17.78	20	22.22	14	18.67	103	18.01	44	26.67	22	18.64	7	21.21
Midwest	42	18.58	21	46.67	20	22.22	22	29.33	118	20.63	50	30.30	29	24.58	9	27.27
South	88	38.94	8	17.78	36	40.00	23	30.67	254	44.41	33	20.00	44	37.29	12	36.36
West	45	19.91	8	17.78	14	15.56	15	20.00	96	16.78	37	22.42	23	19.49	5	15.15
Unknown	1	0.44	0	0.00	0	0.00	1	1.33	1	0.17	1	0.61	0	0.00	0	0.00
Diagnosis
Leukemia	138	61.06	24	53.33	68	75.56	45	60.00	24	4.20	7	4.24	3	2.54	3	9.09
MDS	19	8.41	10	22.22	4	4.44	12	16.00	1	0.17	1	0.61	0	0.00	0	0.00
Lymphoma	49	21.68	6	13.33	15	16.67	13	17.33	246	43.01	59	35.76	48	40.68	9	27.27
Myeloma	14	6.19	5	11.11	2	2.22	2	2.67	288	50.35	97	58.79	64	54.24	18	54.55
Others	6	2.65	0	0.00	1	1.11	3	4.00	13	2.27	1	0.61	3	2.54	3	9.09
NCI CCI
0	20	8.85	3	6.67	18	20.00	7	9.33	53	9.27	9	5.45	8	6.78	3	9.09
1–2	120	53.10	20	44.44	31	34.44	39	52.00	287	50.17	66	40.00	49	41.53	9	27.27
3+	86	38.05	22	48.89	41	45.56	29	38.67	232	40.56	90	54.55	61	51.69	21	63.64
Diabetes	27	11.95	9	20.00	20	22.22	12	16.00	74	12.94	23	13.94	23	19.49	9	27.27
Hypertension	68	30.09	13	28.89	33	36.67	21	28.00	221	38.64	79	47.88	53	44.92	18	54.55
Dyslipidemia	47	20.80	16	35.56	25	27.78	13	17.33	123	21.50	52	31.52	24	20.34	8	24.24
Transfusion*	45	19.91	7	15.56	14	15.56	22	29.33	57	9.97	16	9.70	12	10.17	5	15.15
Mucositis*	62	27.43	14	31.11	23	25.56	19	25.33	126	22.03	47	28.48	24	20.34	5	15.15
Opportunistic Infection*	47	20.80	14	31.11	16	17.78	16	21.33	25	4.37	9	5.45	6	5.08	3	9.09
GVHD*	83	36.73	19	42.22	30	33.33	37	49.33

^*Assessed over the 101–365 day post-transplant period; HSCT, hematopoietic cell transplantation; MDS, Myelodysplastic syndrome; NCI, National Cancer Institute; CCI, Charlson Comorbidity Index; GVHD, Graft-Versus-Host Disease.

Table 2:

Subgroup characteristics in health service uses and costs patterns

		Class identification period (101–365 day post-index)					Class identification validation (366–730 day post-index)
	Service types	Class 1 OPT Specialist	Class 2 OPT PCP	Class 3 OPT/IPT Balanced	Class 4 IPT dominant	p-values	Class 1 OPT Specialist	Class 2 OPT PCP	Class 3 OPT/IPT Balanced	Class 4 IPT dominant	p-values
		(n = 226, 51.8%)	(n = 45, 10.3%)	(n = 90, 20.6%)	(n = 75, 17.2%)		(n = 226, 51.8%)	(n = 45, 10.3%)	(n = 90, 20.6%)	(n = 75, 17.2%)
Allogeneic HSCT	Total Hospital admission*	0.00	0.02	1.17	3.33	<.0001	0.51	0.67	0.92	1.53	<.0001
	Monthly Hospital stays (in days) *	0.00	0.03	0.55	3.67	<.0001	0.30	0.33	0.51	1.22	<.0001
	Total ER visits	0.37	0.49	1.09	3.11	<.0001	0.92	1.40	1.37	1.95	0.0953
	Monthly PCP visits*	0.57	2.02	0.75	1.79	<.0001	0.54	1.55	0.61	1.12	<.0001
	Monthly Specialist visits*	1.29	0.04	1.54	1.56	<.0001	0.78	0.08	0.80	0.78	0.0001
	Mean monthly inpatient costs	88	187	3,227	26,063	<.0001	1,615	2,116	2,284	6,834	<.0001
	Mean monthly outpatient costs	5,141	4,361	6,727	12,985	<.0001	4,198	3,637	4,299	6925	0.0451
	Mean monthly prescription drugs costs	1,383	1,551	1,597	2,049	0.0839	1,138	1,137	1,024	1,297	0.8693
	Mean monthly total costs	6,612	6,099	11,551	41,097	<.0001	6,951	6,890	7,608	15,057	<.0001
		Class 1 OPT Specialist	Class 2 OPT PCP	Class 3 OPT/IPT Balanced	Class 4 IPT dominant	p-values	Class 1 OPT Specialist	Class 2 OPT PCP	Class 3 OPT/IPT Balanced	Class 4 IPT dominant	p-values
		(n = 572, 57.3%)	(n = 165, 25.7%)	(n = 118, 13.5%)	(n = 33, 3.5%)		(n = 572, 57.3%)	(n = 165, 25.7%)	(n = 118, 13.5%)	(n = 33, 3.5%)
Autologous HSCT	Total Hospital admission*	0.00	0.00	1.05	3.06	<.0001	0.30	0.28	0.82	1.73	<.0001
	Monthly Hospital stays (in days) *	0.00	0.00	0.71	2.34	<.0001	0.16	0.14	0.56	1.01	<.0001
	Total ER visits	0.31	0.15	1.07	2.52	<.0001	0.67	0.61	1.18	1.94	0.0058
	Monthly PCP visits*	0.36	1.33	0.66	2.14	<.0001	0.43	1.27	0.63	1.21	<.0001
	Monthly Specialist visits*	0.89	0.13	1.13	0.99	<.0001	0.84	0.20	0.89	0.95	<.0001
	Mean monthly inpatient costs	23	0	3,659	13,423	<.0001	1,061	589	2,494	4,323	0.0002
	Mean monthly outpatient costs	2,954	3,196	5,023	10,178	<.0001	3,364	3,990	4,464	9,988	<.0001
	Mean monthly prescription drugs costs	2,199	2,900	2,067	1,955	0.1175	2,489	2,918	2,131	2,958	0.3534
	Mean monthly total costs	5,175	6,096	10,749	25,556	<.0001	6,914	7,497	9,089	17,269	<.0001

In 2014 US dollar (allogeneic HSCT n=436, autologous HSCT n=888); HSCT, hematopoietic cell transplantation; OPT, outpatient; PCP, Primary care provider, ER emergency room

Results from the logistic regression models evaluating characteristics associated with high-need class membership versus other identified classes are reported in Table 3. We adjusted for empirically selected covariates that included age and gender. After multivariable adjustment, presence of GVHD (OR=1.76, 95% CI 1.05–2.94) and transfusion-related events (OR=1.87, 95% CI 1.06–3.30) during the 100-day post-transplant period were associated with high-utilization class in the allogeneic HSCT cohort over the first year post-transplant. Higher pre-transplantation baseline NCI-CCI was associated with high-utilization class in the autologous HSCT cohort (OR=1.45, 95% CI 1.19–1.76). No other transplant-related characteristics were found to be associated with high healthcare services utilization in the autologous HSCT cohort.

Table 3.

Multivariate logistic regression based on whether the subject belonged to high-user class category (high-need users vs. outpatient service dominant / inpatient outpatient balanced users/ vs), adjusted for age and gender.

Allogeneic HSCT
	Odds ratio	95%CI		p-value
Baseline NCI CCI	0.97	0.78	1.21	0.7931
Transfusion	1.87	1.06	3.30	0.0316
Mucositis	0.84	0.46	1.52	0.5576
Opportunistic Infection	0.91	0.49	1.69	0.7573
GVHD	1.76	1.05	2.94	0.0316
Autologous HSCT
	Odds ratio	95%CI		p-value
Baseline NCI CCI	1.45	1.19	1.76	0.0002
Transfusion	1.72	0.63	4.66	0.2891
Mucositis	0.53	0.20	1.40	0.1995
Opportunistic Infection	2.08	0.60	7.29	0.2513

HSCT, hematopoietic cell transplantation; NCI, National Cancer Institute; CCI, Charlson Comorbidity Index; GVHD, graft-versus-host disease.

In a sensitivity analysis, despite high entropy score, we additionally employed the logistic regression models weighted by individuals’ maximum class membership posterior probability to assess predictors for high-need healthcare service users (latent class 4), the analyses yielded similar results (data not shown).

Healthcare service utilization patterns and healthcare related expenditures over the 366–730 day post-transplant were compared across latent classes to assess the validity of latent class membership. We found significantly higher inpatient admission rates in addition to hospital stays in the high-need users (mean 1.53 admission events per year and 1.22 days of monthly hospital stay in allogeneic HSCT cohort, p<0.0001; mean 1.73 admission events per year and 1.01 days of monthly hospital stay in allogeneic HSCT cohort, p<0.0001). We also observed higher outpatient service use among the inpatient dominant subgroup, and no statistically significant differences in ER visits were found in both the allogeneic and autologous HSCT cohorts.

During the 365–730 day post-transplant period, mean monthly costs per patient were $15,057, $7,608, $6,951 and $6,890 in the IPT dominant, OPT/IPT balanced, specialist care dominant and primary care dominant subgroups in the allogeneic HSCT cohort (p<0.0001). The mean monthly expenditures per patient in the IPT dominant, OPT/IPT balanced, specialist care dominant and primary care dominant subgroups were $17,269, $9,089, $6,914 and $7,497 in the autologous HSCT cohort (p<0.001) The mean monthly costs were lower compared to the first year, although significant differences remained across latent classes. (Table 2). Differences in healthcare spending were driven by inpatient and outpatient services, no statistically significant differences were observed in prescription drugs spending during the second year post-transplant.

Discussion

To our knowledge, this is the first real-world study using latent class analysis to identify classes of adult HSCT recipients’ healthcare service utilization patterns after undergoing HSCT. We found four clinically meaningful healthcare services use patterns that were associated with distinct needs in the post-transplantation long-term follow-up period. Furthermore, the presence of GVHD and transfusion were associated with high need for health service among long-term allogeneic HSCT survivors. In contrast, only higher pre-transplantation baseline NCI CCI was associated with high-need and high-cost of healthcare services among autologous HSCT long-term survivors.

Unlike other analytical approaches, latent class analysis enables researchers to use observed health care use variables and classify observations into subgroups. The approach allows for consideration of multiple outcomes simultaneously and offers an alternative way to identify and describe data. In our study, we identified a small subset of high-use and high-cost patients with long-term survivorship post-transplantation. The differences in health care use and economic burdens were consistent through the second year post-transplant. Knowledge on the anticipated care and healthcare related services costs of HSCT recipients should be incorporated into survivorship care planning by clinical providers and health coverage design by insurance providers and third-party payers.

We observed several factors associated with high-need and high-cost healthcare use. Occurrence of GVHD is known to be an important indicator of early post-transplantation complications, recent wider adoption of peripheral-blood stem cells (PBSC) as the source of allogeneic HSCT has contributed to increased risks of GVHD within the 100-day post-transplant period [24, 25]. GVHD could affect skin, liver and gastrointestinal tract and cause substantial morbidities and mortality. Due to the limitations in our data, we were unable to differentiate classic acute GVHD and overlap chronic GVHD syndrome, or to comprehensively evaluate the clinical involvement in addition to the severity of GVHD among these patients. Moreover, we were unable to ascertain whether these patients responded to treatment for GVHD. Requirement for transfusion was also associated with high-need, high-cost latent class membership. Indeed, patients experiencing graft rejection or delayed hematological reconstitution following allogeneic HSCT may be at greater risk of red cell and platelet transfusion dependence [26]. Early identification of high-need patients is essential to managing complex cases requiring multi-disciplinary care. Proactive provision of tailored care has the potential to mitigate the impact of transplant-related life-threatening complications, restore patient functional and psychological wellbeing and reduce future healthcare spending.

Our study also highlights the considerable burdens and significant difference between patients with distinctive post-transplantation care patterns. Another study of 1141 allogeneic HSCT patients found that 30-day readmission rates were 28.3% and 17.5% in myeloablative and reduced-intensity conditioning HSCT respectively [27]. The 100-day readmission rates were 42.8% and 31.1% respectively with infection being the most frequent reason and a significant risk factor for readmission [27]. The latent class memberships were found to have strong associations with post-transplantation costs and service use. Multi-level interventions, such as close monitoring of patients’ immunosuppressive state, careful tapering of immunosuppressive therapy, and early prevention and management of infections in the identified high need patients at day 100 post-transplant may reduce the overall health care costs post-transplant while optimizing clinical outcomes.

Our study found that inpatient service was the main driver for higher per patient per month (PPPM) costs in the post-transplant period, which is similar to other studies. Another real-world economic burden study in HSCT survivors found that inpatient admissions accounted for 75.5% and 54.1% of all-cause health costs in allogeneic and autologous HSCT recipients during the 18-month post-transplantation period [28]. In our study, we identified a group of patients with costs two- to five-fold higher that were driven largely by higher rates of inpatient service utilization. Occurrence of GVHD and transfusion were also important contributors to this pattern. Interestingly, the inpatient dominant subgroup identified from this population-based sample not only showed substantial higher utilization rates of inpatient services but also greater ER visits and primary care encounters. These findings suggest that prompt transfer of patients to outpatient settings through reduction of inpatient length of stay could be part of the solution. Better understanding of the coordination of complex health conditions management through multi-disciplinary interventions approach are essential for optimizing care for high-need, high-cost HSCT recipients.

A notable strength of our study is that we used a large, administrative database that covers more than 100 million commercially insured patients and their dependents. We were able to include objectively collected accurate information on healthcare related service use data from a national representative population. Unlike other studies that used survey questionnaires and self-reported outcomes [29], our longitudinal data ensured continuous follow-up with objectively collected outcomes even if patients were transferred from the transplant center where HSCT was received to other local healthcare facilities. We were able to collect demographic and clinical factors that could influence the health services utilization patterns. Moreover, our study included inpatient, outpatient and prescription drug costs to assess the comprehensive economic burdens to the HSCT recipients.

Our findings add to the existing literature on survivorship of HSCT recipients in several ways. First, our study included patients undergoing HSCT across various malignant indications, while other similar published work was limited to acute myeloid leukemia [30, 31]. Second, our analyses included contemporary data that were representative of recent treatment patterns and indications for HSCT. Lastly, this is the first study demonstrating feasibility of LCA application in clustering HSCT recipients and provide a quantitative evaluation of healthcare needs and medical costs from a payer’s perspective. Commercially-insured patients face risk financial toxicity from the substantial medical and nonmedical costs incurred before, during and after HSCT, possibly resulting worse outcomes [32]. We found from our secondary analyses that healthcare use and medical costs patterns persisted through second year among long-term HSCT survivors. Another study of 320 allogeneic and autologous HSCT survivors found that fatigue and sexual difficulties were reported in more than 24% patients after transplant [33]. The patients in the high utilization group were distinct because their underlying illness required constant care and frequent specialized care visits, which may be not available in the proximity of their original residential regions. Future investigation is needed to examine the financial burden, persistent symptom management and issues with access to quality care post-HSCT among cancer survivors, particularly from the perspective of patients and their families.

There are also some limitations to our study worth mentioning. First, we focused on patients who were enrolled in private insurance and were long-term survivors. The findings might not be generalizable to patients who were publicly insured and had less than two years’ survival. To heighten the external validity, future studies should include publicly insured HSCT populations. Second, in the absence of HSCT clinical information including donor source, CMV status, ABO/HLA matching status, etc., we employed surrogate endpoints (transfusion events in lieu of hematological recovery data). However, the use of this administrative database enables us to gain access to a broad range of patients with wide variation in demographic and clinical characteristics who would not necessarily be otherwise eligible for clinical trials. The real-world nature of our data allowed for a more nuanced understanding of the care pathway following transplant and this information can help inform care coordination among long-term HSCT survivors. Third, while we created mutually exclusive categories to differentiate specialty and primary care using provider types, the healthcare services and associated costs may be underestimated if the provider types were missing or not included in analyses. Lastly, as in any observational study using an administrative claims database, unmeasured confounding and misclassification is always possible. However, per validated algorithms, we required patients to meet a number of inclusion and exclusion criteria to enter the analysis and minimized some of these biases inherent to the data source.

Conclusions

The present study demonstrates that long-term adult HSCT survivors consist of subgroups characterized by distinct healthcare services use patterns. From a clinical perspective, we validated the discriminatory power of latent classes in distinguishing high-need, high-cost users throughout the two-year follow-up period. We further identified factors associated with high-need, high-cost healthcare utilization patterns. Such knowledge calls for a personalized case management strategy. Additional research and approaches to allow for early identification of patients with complex post-transplantation complications to address high disease burden and financial distress are warranted.