Financial Toxicity Profiles and Associated Factors in Patients with Advanced Cancer Receiving Daytime Chemotherapy: A Latent Profile Analysis

Abstract

Background

Financial toxicity (FT) may persist among patients with advanced cancer receiving daytime chemotherapy because of prolonged treatment, frequent hospital visits, and nonmedical expenses. However, heterogeneity in this outpatient setting remains insufficiently understood.

Aim

To identify latent FT profiles among patients with advanced cancer receiving daytime chemotherapy and examine factors associated with profile membership.

Methods

This cross-sectional study recruited 335 patients with advanced cancer receiving daytime chemotherapy in Southwest China from November 2025 to January 2026. Participants completed the COST-PROM scale, Hospital Anxiety and Depression Scale, and Family APGAR Index. Latent profile analysis identified FT profiles, and forced-entry multinomial logistic regression examined adjusted associations.

Results

Three profiles were identified: High FT (32.5%, n=109), Moderate FT (31.7%, n=106), and Low FT (35.8%, n=120). Univariate comparisons showed significant differences across demographic, socioeconomic, clinical, and psychosocial characteristics. With High FT as the reference, a higher total HADS score, representing overall psychological distress, was associated with higher odds of Moderate FT membership (adjusted odds ratio (aOR) = 1.081, 95% CI: 1.042–1.121), whereas disease duration < 6 months versus > 1 year was associated with lower odds (aOR = 0.437, 95% CI: 0.203–0.937). For Low versus High FT membership, disease duration <6 months versus >1 year (aOR = 4.394, 95% CI: 1.749–11.041) and each one-point increase in Family APGAR score (aOR = 1.536, 95% CI: 1.235–1.910) were associated with higher adjusted odds.

Conclusion

FT showed cross-sectional heterogeneity in this population. The observed associations may inform context-specific assessment and supportive-care planning, but profile selection and several regression estimates remain uncertain. External and longitudinal validation is required.

Introduction

Cancer remains one of the leading causes of death worldwide and is a major contributor to global disease burden, exerting substantial economic and social pressure on public health systems.¹ Due to its insidious onset and lack of early symptoms, many patients are diagnosed at advanced stages.² Consequently, patients with advanced cancer constitute a large proportion of those receiving antineoplastic treatment. Chemotherapy is one of the most widely used and effective treatment modalities and has been shown to significantly prolong survival.^3,4 However, patients with advanced cancer often experience severe symptoms and related side effects, causing a substantial psychological and economic burden on both the patients and their families.⁵ Moreover, as the disease progresses to its advanced stages, treatment costs tend to increase further.⁶ This phenomenon highlights the significant financial challenges faced by patients with advanced cancer, particularly those undergoing chemotherapy.

In this context, financial toxicity (FT) has emerged as a critical issue in cancer care. FT refers to the overall negative impact of cancer-related costs on patients and their families, encompassing both objective financial burden (eg, direct and indirect medical expenses) and subjective financial distress (eg, stress, anxiety, and emotional strain).⁷ Previous studies have reported that the prevalence of FT among cancer patients ranges from 48% to 73%, and severe FT may lead patients to modify or even discontinue treatment, thereby adversely affecting treatment adherence and quality of life. Although health insurance coverage for anticancer drugs has expanded in China, many patients still face high out-of-pocket expenditures due to socioeconomic disparities and policy limitations.^8,9

Daytime chemotherapy has emerged as an alternative to traditional inpatient chemotherapy for patients with relatively stable conditions. This model allows patients to complete examinations and chemotherapy within a single day and return home the same day.¹⁰ Compared with inpatient chemotherapy, daytime chemotherapy reduces hospitalization time, bed occupancy, and accommodation costs, thereby alleviating part of the economic burden. However, patients undergoing daytime chemotherapy may still experience high FT due to long-term medication costs, frequent transportation, and increased caregiving needs.^11,12 Therefore, understanding heterogeneity in FT and the characteristics associated with profile membership among patients with advanced cancer receiving daytime chemotherapy is clinically relevant.

Previous studies have examined FT among general cancer populations and patients with specific cancer types, including those receiving conventional inpatient treatment. Latent profile analysis has also been used to identify heterogeneous patterns of FT and their associated characteristics, and associations between FT and socioeconomic circumstances, disease-related characteristics, psychological distress, and family support have been documented in previous cancer populations.^7,13 FT and psychological distress may have reciprocal relationships, while family functioning may be associated with both financial and psychological experiences. These relationships are therefore considered without assuming a causal direction in the present cross-sectional study. However, existing evidence has mainly been derived from general cancer populations or conventional treatment settings, whereas patients with advanced cancer receiving daytime chemotherapy remain underrepresented. Daytime chemotherapy differs from conventional inpatient chemotherapy in that it may reduce hospitalization-related costs while shifting part of the financial burden to repeated outpatient visits and associated transportation, medication, caregiving, and time-related costs. Therefore, it remains unclear how previously identified patterns of FT manifest within this specific treatment setting and whether profile membership is associated with demographic, clinical, and psychosocial characteristics among patients with advanced cancer receiving daytime chemotherapy.

Identifying clinically meaningful FT profiles among patients with advanced cancer receiving daytime chemotherapy may provide a context-specific basis for financial risk assessment and supportive care planning. Latent profile analysis (LPA), an individual-centered analytical approach, enables the identification of unobserved subgroups based on patients’ response patterns and is therefore appropriate for exploring heterogeneous experiences of FT.¹⁴ Unlike traditional variable-centered approaches, LPA can capture complex patterns of co-occurring responses and summarize heterogeneity at the individual level.¹⁵ Accordingly, this study aimed to characterize FT profiles among patients with advanced cancer receiving daytime chemotherapy and examine demographic, socioeconomic, clinical, and psychosocial characteristics associated with profile membership. The findings were intended to inform context-specific assessment and supportive-care planning.

Methods

Participants and Setting

A cross-sectional study was conducted using convenience sampling to recruit patients with advanced cancer receiving chemotherapy in daytime chemotherapy wards at three tertiary hospitals in Chengdu, Sichuan Province, between November 2025 and January 2026 (namely, The General Hospital of Western Theater Command, Chengdu First People’s Hospital, and Chengdu Seventh People’s Hospital). The study period represented the recruitment window only; each participant completed the assessment once, and no longitudinal follow-up or repeated FT assessment was conducted. Inclusion criteria were: age ≥ 18 years; pathologically confirmed cancer at TNM stage III or IV; ability to communicate effectively; awareness of the diagnosis; and provision of informed consent. Exclusion criteria included: history of mental illness, severe communication impairment, inability to cooperate due to disease progression, or participation in other interventional studies.

Sample Size Considerations

No formal a priori power or simulation-based sample-size calculation was performed for the final analytical models. Sample adequacy was therefore assessed retrospectively in relation to model complexity, profile sizes, sparse covariate categories, and estimate precision. The final sample comprised 335 participants, with 109, 106, and 120 participants in the retained profiles. Therefore, the adequacy of the sample should be interpreted in terms of model estimation stability and profile interpretability rather than as confirmation of sufficient statistical power. The multinomial logistic regression included eight covariates, generating 28 estimated covariate coefficients across the two logit equations. Although the model was estimable, some profile-by-covariate cells were sparse, limiting the precision of several categorical contrasts.

Procedure and Ethical Considerations

Eligible patients were identified through electronic medical records with the assistance of clinical nurses at each participating hospital. During the study period, all patients who met the eligibility criteria were screened and invited to participate. After receiving a standardized explanation of the study objectives and procedures, patients who agreed to participate provided written informed consent and completed the questionnaires. The recruitment procedures, eligibility assessment, questionnaire administration, and data collection processes were standardized across the three hospitals. Immediately after completion, each questionnaire was reviewed by a trained researcher for completeness and clarity. When missing or unclear responses were identified, participants were invited, where feasible, to complete or clarify their responses before leaving the study site. No responses were inferred or supplied by the researchers. Demographic and clinical information was cross-checked against medical records when available and appropriate. If required information remained unavailable after verification, the questionnaire was considered incomplete and excluded according to the prespecified data-quality criteria. Of the 355 eligible patients approached, 5 were unable to participate because of poor physical condition and 8 declined participation. Of the 342 questionnaires returned, 7 were excluded due to invalid or low-quality responses, leaving 335 participants with valid and complete data in the final analysis (see Figure 1). Among the final sample, 211 participants were recruited from The General Hospital of Western Theater Command, 55 from Chengdu First People’s Hospital, and 69 from Chengdu Seventh People’s Hospital. Following the quality-control procedures, all 335 participants included in the final analysis had complete data required for the primary analyses.

Figure 1.

Flow diagram of participant recruitment and inclusion in the analytical sample. Of the 355 patients approached, five were unable to participate because of poor physical condition and eight declined participation. Of the 342 questionnaires returned, seven were excluded because of invalid or low-quality responses, leaving 335 participants in the final analytical sample.

The study was approved by the Ethics Committee of the General Hospital of the Western Theater Command of the Chinese People’s Liberation Army (Approval No. 2025EC11-ky032) and conducted in accordance with the Declaration of Helsinki.

Measurements

Demographic and clinical characteristics included age, sex, marital status, education level, monthly household income per capita, employment status, residence, disease duration, cancer type, metastatic status, chemotherapy cycles, treatment modality, and medical payment method.

The Comprehensive Score for Financial Toxicity Based on Patient-Reported Outcome Measures (COST-PROM) Scale

The Chinese version of the COST-PROM Scale, which was translated and culturally adapted for Chinese cancer patients, was used in this study.¹⁶ The 11 items included in the analysis were scored from 0 to 4. Items 2, 3, 4, 5, 8, 9, and 10 were reverse-scored according to the scoring instructions before analysis. All items entered into the LPA were oriented consistently, with lower scores indicating greater FT. In this study, the Cronbach’s α coefficient of the scale was 0.927.

Hospital Anxiety and Depression Scale (HADS)

The Chinese version of the HADS, which has demonstrated satisfactory reliability and validity among Chinese populations, was used in this study.¹⁷ The scale consists of 14 items and includes two subscales: the Anxiety subscale (HADS-A) and the Depression subscale (HADS-D). The 14 item scores were summed to obtain a total HADS score ranging from 0 to 42. The total score was entered into the multinomial logistic regression as a single continuous covariate representing overall psychological distress. It was not used to estimate the independent associations of the anxiety and depression subscales. The Cronbach’s α coefficient of the scale in this study was 0.962.

Family APGAR Index (APGAR)

The Chinese version of the Family APGAR Index, which has been widely applied to evaluate family functioning among Chinese populations, was used in this study.¹⁸ The APGAR includes five dimensions: Adaptation, Partnership, Growth, Affection, and Resolve. The total score ranges from 0 to 10, with higher scores indicating better family functioning. In this study, the Cronbach’s α coefficient of the scale was 0.869.

Statistical Analysis

Data were analyzed using IBM SPSS Statistics version 27.0 and Mplus version 8.3. Data completeness was checked before analysis. The final analytical dataset contained no missing values for the 11 COST-PROM indicators or the covariates included in the reported analyses. All analyses used the same 335 participants, and no imputation was performed. Continuous variables were summarized as means and standard deviations or medians and interquartile ranges, as appropriate; categorical variables were summarized as frequencies and percentages.

LPA was conducted using the 11 COST-PROM items to identify FT profiles among patients with advanced cancer receiving daytime chemotherapy. The five-point Likert items were treated as continuous indicators, and models were estimated using robust maximum likelihood (MLR). This specification approximated the ordered responses as continuous; sensitivity to an ordered categorical specification was not evaluated. Item means were freely estimated across profiles, item variances were constrained to be equal across profiles, and within-profile covariances were fixed at zero. To reduce the risk of local maxima, 200 random initial-stage starts and 50 final-stage optimizations were specified, and replication of the best log-likelihood value was confirmed. Models with one to five profiles were compared using log-likelihood values, the Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size-adjusted BIC (aBIC), entropy, the Lo–Mendell–Rubin adjusted likelihood-ratio test (LMR), and the bootstrap likelihood-ratio test (BLRT). Lower information criterion values indicated better relative fit after accounting for model complexity, whereas higher entropy indicated better classification quality.¹⁹ Significant LMR and BLRT results supported the k-profile model over the corresponding k−1-profile model.²⁰ The retained solution was selected by jointly considering statistical fit, classification quality, profile sizes, parsimony, and substantive interpretability.

For the retained solution, participants were assigned to the profile with the highest posterior probability. These most-likely assignments were treated as an observed nominal outcome in subsequent unadjusted comparisons and multinomial logistic regression analyses. Consequently, the regression estimates did not account for posterior classification uncertainty. Unadjusted differences in demographic, socioeconomic, clinical, and psychosocial characteristics across profiles were examined using chi-square tests, Fisher’s exact tests, one-way analysis of variance, or Kruskal–Wallis tests, as appropriate. Covariates for the multinomial logistic regression were selected with reference to previous evidence, clinical relevance, and the study objectives rather than solely on the basis of unadjusted statistical significance. Eight covariates were entered simultaneously using a forced-entry approach: age, sex, monthly household income per capita, employment status, medical payment method, disease duration, total HADS score, and Family APGAR score. These represented demographic characteristics, socioeconomic and healthcare-financing circumstances, time since diagnosis, overall psychological distress, and family functioning. Education level and residence were not included in the adjustment set to limit model complexity. This decision reflected their conceptual relationship with the selected socioeconomic covariates, but did not imply that income, employment status, and medical payment method fully captured their associations with FT profile membership. Residual confounding related to education and residence therefore remains possible.

Cancer type was also considered clinically relevant because differences in prognosis, treatment intensity, costs, and symptom burden could potentially confound associations with FT profile membership. However, the five-category cancer-type variable would have added four covariate coefficients to each logit equation, increasing the total number of covariate coefficients from 28 to 36. Because some profile-by-cancer-type cells contained relatively few participants, cancer type was not included in the primary model to limit overparameterization and unstable estimation. This decision was based on clinical consideration, data support, and model parsimony rather than on its univariate P value alone.

The High FT Group was specified as the reference outcome. Accordingly, each adjusted odds ratio (aOR) represented the odds of membership in the Moderate or Low FT Group versus the High FT Group after adjustment for all other covariates included in the model. An aOR >1 indicated higher odds, whereas an aOR < 1 indicated lower odds of membership in the specified comparison profile. For continuous variables, the aOR reflected the change in odds associated with a one-point increase; for categorical variables, it represented comparisons between each category and its prespecified reference. Linearity in the logits for the continuous covariates (total HADS score and Family APGAR score) was assessed using Box–Tidwell transformation terms. Multicollinearity was assessed using an auxiliary linear regression containing the same 14-column predictor matrix as the final multinomial logistic regression. Multicategorical covariates were represented by indicator variables using prespecified reference categories. Tolerance values and variance inflation factors (VIFs) were examined; the coefficients and significance tests from the auxiliary regression were not interpreted.

Profile-by-covariate cross-tabulations were examined for sparse cells, and model stability was assessed based on convergence, parameter estimates, confidence interval widths, and separation warnings. Although sparse category merging was considered, the original categories were retained because they were prespecified and clinically and socioeconomically meaningful. The conventional multinomial logistic regression model was retained because all parameters were estimable and no separation or convergence issues were identified; however, estimates with wide confidence intervals were interpreted cautiously. Model performance was evaluated using the likelihood-ratio test, Pearson and deviance goodness-of-fit tests, and Cox–Snell, Nagelkerke, and McFadden pseudo-R² statistics. Statistical significance was set at P < 0.05.

Results

After data-quality and completeness checks, 335 participants with complete data for all analytical variables were included. Most were male (66.0%), aged 18–64 years (60.9%), and married (83.0%). Overall, 39.4% had primary school education or below, 9.6% had a college education or above, and 40.0% were classified as not working. Participants resided in cities (42.7%), towns (29.0%), and rural areas (28.3%); 12.2% reported a monthly household income per capita exceeding 5000 CNY. Gastrointestinal cancers were the most common cancer category (39.4%). Most participants had metastatic disease (76.1%) and received chemotherapy combined with other treatment modalities (89.6%). Additional characteristics are presented in Supplementary Material 1.

Models with one to five profiles were compared (Table 1). AIC, BIC, and aBIC decreased as profiles were added, and the LMR and BLRT remained significant for the four- and five-profile solutions. The five-profile solution also had slightly higher entropy than the three-profile solution (0.944 versus 0.941). Thus, statistical criteria did not uniquely favor three profiles. The three-profile solution was retained as a parsimonious representation of relatively higher, intermediate, and lower FT patterns, whereas the four- and five-profile solutions captured additional item-specific variation (Supplementary Material 2). This selection represents an interpretive choice rather than the only plausible classification.

Table 1.

Fit Indices for the Five Models Using LPA (N = 335)

Profile	AIC	BIC	aBIC	Entropy	LMR (P)	BLRT (P)	Profile Prevalence
1	11,423.756	11,507.667	11,437.881	—	—	—	1.00
2	9686.598	9816.278	9708.426	0.950	< 0.001	< 0.001	0.534/0.466
3	9203.749	9379.199	9233.283	0.941	< 0.001	< 0.001	0.325/0.317/0.358
4	8806.175	9027.394	8843.412	0.932	0.0340	< 0.001	0.224/0.281/0.257/0.239
5	8567.075	8834.064	8612.017	0.944	0.0228	< 0.001	0.188/0.185/0.272/0.104/0.251

Notes: Profile prevalence is reported as the proportion of participants in each profile. —, not applicable.

Abbreviations: LPA, latent profile analysis; AIC, Akaike information criterion; BIC, Bayesian information criterion; aBIC, sample-size-adjusted Bayesian information criterion; LMR, Lo–Mendell–Rubin adjusted likelihood-ratio test; BLRT, bootstrap likelihood-ratio test.

Based on most-likely profile assignment, the three profiles comprised 109 participants with relatively high FT (32.5%), 106 with moderate FT (31.7%), and 120 with relatively low FT (35.8%). Figure 2 illustrates their response patterns across the 11 COST-PROM indicators. Because lower consistently coded item scores indicate greater FT, the profiles with the lowest, intermediate, and highest scores were labelled the High FT, Moderate FT, and Low FT Groups, respectively. These labels describe relative differences within the study sample rather than established clinical severity thresholds.

Figure 2.

Mean COST-PROM item scores across the three retained latent financial toxicity profiles. The High FT, Moderate FT, and Low FT profiles comprised 109 (32.5%), 106 (31.7%), and 120 (35.8%) participants, respectively. FT1–FT11 denote the 11 COST-PROM items included in the latent profile analysis. Item scores were coded consistently so that lower scores indicate greater financial toxicity.

Abbreviations: FT, financial toxicity; COST-PROM, Comprehensive Score for Financial Toxicity.

Univariate analyses were conducted to compare demographic, clinical, and psychosocial characteristics across the three latent FT profiles among patients with advanced cancer receiving daytime chemotherapy. Significant between-profile differences were observed in sex, education, monthly household income per capita, residence, medical payment method, employment status, disease duration, chemotherapy cycles, overall psychological distress, and family functioning (see Table 2).

Table 2.

Univariate Comparisons of Demographic, Socioeconomic, Clinical, and Psychosocial Characteristics Across Financial Toxicity Profiles (N = 335)

Variables	Categories	High FT Group n (%)	Moderate FT Group n (%)	Low FT Group n (%)	Statistics	P value
Demographic and socioeconomic characteristics
Sex	Male	71 (65.10)	56 (52.80)	94 (78.30)	16.356a	< 0.001
	Female	38 (34.90)	50 (47.20)	26 (21.70)
Age	18~64 years	67 (61.50)	72 (67.90)	65 (54.20)	4.496a	0.106
	≥65 years	42 (38.50)	34 (32.10)	55 (45.80)
Marital status	Married	87 (79.80)	90 (84.90)	101 (84.20)	1.171a	0.557
	Unmarried/Divorce/Widowed	22 (20.20)	16 (15.10)	19 (15.80)
Educational level	Primary school or below	51 (46.80)	45 (42.50)	36 (30.00)	52.758a	< 0.001
	Junior high school	43 (39.40)	45 (42.50)	25 (20.80)
	High school or vocational school	10 (9.20)	14 (13.10)	34 (28.40)
	College or above	5 (4.60)	2 (1.90)	25 (20.80)
Average monthly household income per capita (CNY)	≤ 3000	59 (54.10)	72 (67.90)	12 (10.00)	112.112a	< 0.001
	3001~5000	47 (43.10)	33 (31.20)	71 (59.20)
	>5000	3 (2.80)	1 (0.90)	37 (30.80)
Employment status	Working	6 (5.50)	9 (8.50)	17 (14.20)	88.980a	< 0.001
	Not working	58 (53.20)	44 (41.50)	32 (26.70)
	Peasants	35 (32.10)	50 (47.20)	19 (15.80)
	Retired	10 (9.20)	3 (2.80)	52 (43.30)
Residence	City	42 (38.50)	30 (28.30)	71 (59.20)	25.378a	< 0.001
	Town	34 (31.20)	34 (32.10)	29 (24.20)
	Countryside	33 (30.30)	42 (39.60)	20 (16.60)
Primary caregiver	None	20 (18.30)	12 (11.30)	20 (16.70)	6.108a	0.411
	Mate	55 (50.50)	68 (64.20)	73 (60.80)
	Child	32 (29.40)	24 (22.60)	24 (20.00)
	Others	2 (1.80)	2 (1.90)	3 (2.50)
Number of children	0	3 (2.80)	4 (3.80)	3 (2.50)	5.993b	0.424
	1	52 (47.70)	55 (51.90)	66 (55.00)
	2	47 (43.10)	43 (40.50)	39 (32.50)
	≥ 3	7 (6.40)	4 (3.80)	12 (10.00)
Payment method for medical expenses	URBMI	29 (26.60)	20 (18.80)	35 (29.20)	75.907a	< 0.001
	NRCMS	58 (53.20)	68 (64.20)	27 (22.50)
	UEBMI	13 (11.90)	9 (8.50)	56 (46.70)
	Self-paid	9 (8.30)	9 (8.50)	2 (1.60)
Clinical characteristics
Types of cancer	Lung Cancer	37 (33.90)	26 (24.50)	37 (30.80)	12.200a	0.142
	Gastrointestinal Cancer	34 (31.20)	43 (40.60)	55 (45.80)
	Head and Neck Cancer	17 (15.60)	20 (18.90)	13 (10.80)
	Gynecological Cancer	17 (15.60)	13 (12.20)	8 (6.80)
	Other types of Cancer	4 (3.70)	4 (3.80)	7 (5.80)
Tumour metastasis	Yes	82 (75.20)	79 (74.50)	94 (78.30)	0.519a	0.772
	No	27 (24.80)	27 (25.50)	26 (21.70)
Synchronize with other treatment methods	No	12 (11.00)	11 (10.40)	12 (10.00)	0.063a	0.969
	Yes	97 (89.00)	95 (89.60)	108 (90.00)
Disease duration	Less than six months	36 (33.00)	23 (21.70)	62 (51.70)	37.235a	< 0.001
	Six months to one year	35 (32.10)	28 (26.40)	39 (32.50)
	More than one year	38 (34.90)	55 (51.90)	19 (15.80)
Chronic comorbidities	0	63 (57.80)	66 (62.30)	72 (60.00)	1.667a	0.797
	1	37 (33.90)	35 (33.00)	42 (35.00)
	≥2	9 (8.30)	5 (4.70)	6 (5.00)
Round of chemotherapy	1~4	44 (40.40)	17 (16.10)	57 (47.50)	26.316a	< 0.001
	5~8	36 (33.00)	49 (46.20)	34 (28.30)
	≥9	29 (26.60)	40 (37.70)	29 (24.20)
Psychosocial characteristics
Overall psychological distress (anxiety and depressive symptom burden)		12.00 (6.00, 19.50)	22.00 (14.00, 26.25)	11.00 (6.00, 16.00)	47.169c	< 0.001
Family functioning (Family APGAR)		9.00 (6.00, 10.00)	7.00 (5.00, 9.00)	10.00 (8.25, 10.00)	56.485c	< 0.001

Notes: Categorical variables are presented as n (%), with percentages calculated within each FT profile. Continuous variables are presented as median (Q1, Q3), where Q1 and Q3 denote the 25th and 75th percentiles, respectively. Bold section headings identify domains of characteristics and do not indicate statistical significance. Superscripts indicate the statistical tests used: a, Pearson’s chi-square test; b, Fisher’s exact test; c, Kruskal–Wallis test. P values represent overall comparisons across the three profiles.

Abbreviations: FT, financial toxicity; HADS, Hospital Anxiety and Depression Scale; APGAR, Adaptation, Partnership, Growth, Affection, and Resolve; CNY, Chinese yuan; NRCMS, New Rural Cooperative Medical Scheme; UEBMI, Urban Employee Basic Medical Insurance; URBMI, Urban–Rural Resident Basic Medical Insurance.

Before fitting the multinomial logistic regression model, multicollinearity was assessed using an auxiliary linear regression containing the same dummy-coded predictor matrix as the final multinomial model. The tolerance values ranged from 0.759 to 0.930, and the VIFs ranged from 1.075 to 1.318, indicating no evidence of problematic multicollinearity. The Box–Tidwell transformation terms for overall psychological distress and family functioning were not statistically significant (both P > 0.05), providing no evidence of departure from linearity in the logits. Inspection of the profile-by-covariate cross-tabulations indicated that some cells contained relatively few participants, particularly those involving the higher-income, retired, and self-paid categories. The model converged successfully, all parameters were estimable, and no warnings indicating complete or quasi-complete separation were produced. The final model included eight covariates and provided a significantly better fit than the intercept-only model (likelihood-ratio χ² = 256.077, df = 28, P < 0.001). The nonsignificant Pearson and deviance goodness-of-fit tests provided no evidence of inadequate model fit (Pearson χ² = 597.242, df = 632, P = 0.836; deviance χ² = 473.484, df = 632, P = 1.000). These statistics were interpreted cautiously given the presence of sparse profile-by-covariate cells. The Cox–Snell, Nagelkerke, and McFadden pseudo-R² values were 0.534, 0.601, and 0.348, respectively. Detailed model diagnostic and performance measures are presented in Supplementary Material 3.

Adjusted associations with FT profile membership are presented in Table 3, with the High FT Group as the reference outcome. All estimates were mutually adjusted for age, sex, monthly household income per capita, employment status, medical payment method, disease duration, overall psychological distress, and family functioning. For the Moderate FT versus High FT comparison, each one-point increase in the total HADS score was associated with higher adjusted odds of membership in the Moderate FT Group (aOR = 1.081, 95% CI: 1.042–1.121, P < 0.001). A disease duration of less than 6 months, compared with more than 1 year, was associated with lower adjusted odds of membership in the Moderate FT Group (aOR = 0.437, 95% CI: 0.203–0.937, P = 0.033). No other adjusted associations reached statistical significance for this comparison.

Table 3.

Adjusted Multinomial Logistic Regression Analysis of Factors Associated with Financial Toxicity Profile Membership (Reference Outcome: High FT Group; N = 335)

Variable And Category	Moderate FT Group vs High FT Group			Low FT Group vs High FT Group
	β	aOR (95% CI)	P value	β	aOR (95% CI)	P value
Age: 18–64 years (Ref: ≥ 65 years)	0.287	1.332 (0.672–2.639)	0.411	−0.615	0.540 (0.255–1.147)	0.109
Sex: Male (Ref: Female)	−0.167	0.846 (0.452–1.584)	0.602	0.449	1.567 (0.702–3.501)	0.273
Income: ≤ 3000 CNY (Ref: > 5000 CNY)	1.365	3.917 (0.320–47.988)	0.286	−3.563	0.028 (0.005–0.150)	< 0.001
Income: 3001–5000 CNY (Ref: > 5000 CNY)	0.752	2.122 (0.185–24.298)	0.545	−1.809	0.164 (0.038–0.699)	0.015
Employment: Working (Ref: Retired)	1.495	4.458 (0.720–27.618)	0.108	−0.947	0.388 (0.091–1.656)	0.201
Employment: Not working (Ref: Retired)	0.679	1.971 (0.380–10.215)	0.419	−1.123	0.325 (0.107–0.994)	0.049
Employment: Peasants (Ref: Retired)	1.283	3.608 (0.635–20.488)	0.148	−0.813	0.444 (0.116–1.694)	0.234
Payment: URBMI (Ref: Self-paid)	−0.365	0.694 (0.199–2.418)	0.566	0.911	2.486 (0.397–15.571)	0.331
Payment: NRCMS (Ref: Self-paid)	−0.156	0.855 (0.282–2.593)	0.782	0.927	2.526 (0.425–15.018)	0.308
Payment: UEBMI (Ref: Self-paid)	0.080	1.083 (0.226–5.186)	0.920	1.930	6.890 (0.952–49.875)	0.056
Disease duration: < 6 months (Ref: > 1 year)	−0.829	0.437 (0.203–0.937)	0.033	1.480	4.394 (1.749–11.041)	0.002
Disease duration: 6 months–1 year (Ref: > 1 year)	−0.541	0.582 (0.282–1.204)	0.145	1.144	3.138 (1.217–8.092)	0.018
Overall psychological distress	0.078	1.081 (1.042–1.121)	< 0.001	0.007	1.007 (0.965–1.051)	0.757
Family functioning (Family APGAR)	−0.046	0.955 (0.834–1.094)	0.507	0.429	1.536 (1.235–1.910)	< 0.001

Notes: The High FT Group was the reference outcome. All estimates were adjusted for age, sex, monthly household income per capita, employment status, medical payment method, disease duration, overall psychological distress, and family functioning. For the HADS and Family APGAR scores, aORs are reported per one-point increase. Reference categories for categorical covariates are indicated in the table β, regression coefficient.

Abbreviations: aOR, adjusted odds ratio; CI, confidence interval; Ref, reference category; CNY, Chinese yuan; FT, financial toxicity; HADS, Hospital Anxiety and Depression Scale; APGAR, Adaptation, Partnership, Growth, Affection, and Resolve.

For the Low FT versus High FT comparison, each one-point increase in Family APGAR score was associated with higher adjusted odds of Low FT Group (aOR = 1.536, 95% CI: 1.235–1.910, P < 0.001). Compared with monthly household income per capita exceeding 5000 CNY, incomes of 3000 CNY or less (aOR = 0.028, 95% CI: 0.005–0.150, P < 0.001) and of 3001–5000 CNY (aOR = 0.164, 95% CI: 0.038–0.699, P = 0.015) were associated with lower adjusted odds. Participants classified as not working also had lower adjusted odds than retired participants (aOR = 0.325, 95% CI: 0.107–0.994, P = 0.049). Compared with disease duration exceeding 1 year, durations of less than 6 months (aOR = 4.394, 95% CI: 1.749–11.041, P = 0.002) and 6 months to 1 year (aOR = 3.138, 95% CI: 1.217–8.092, P = 0.018) were associated with higher adjusted odds. Age, sex, and medical payment method showed no significant category-specific associations in either comparison. These cross-sectional associations, estimated using most-likely profile assignment, should be interpreted as exploratory.

Discussion

The primary analysis retained a three-profile representation of FT among patients with advanced cancer receiving daytime chemotherapy: High FT (32.5%), Moderate FT (31.7%), and Low FT (35.8%). This severity-based structure is broadly consistent with the three-profile solutions reported by Xiao et al²¹ and Ma et al²² In contrast, Li et al⁹ and Zhang et al²³ identified four profiles, with the moderate group further differentiated by psychological adaptation or resource availability. Such differences may reflect variations in cancer type, disease stage, treatment setting, measurement instruments, and item-response distributions. Within the retained three-profile solution, the consistently ordered, non-crossing COST-PROM response patterns supported an interpretation based on relative FT severity. However, the competing solutions captured additional item-specific heterogeneity, and the three-profile representation should not be interpreted as the only plausible structure.

The contribution of this study lies in providing context-specific evidence for patients with advanced cancer receiving daytime chemotherapy. It extends existing work on FT profiles and associated socioeconomic, disease-related, psychological, and family characteristics to this treatment setting. The study does not introduce a new mixture-modeling method or establish causal relationships. The profiles describe a pooled population comprising several cancer types. Differences in prognosis, treatment regimens, costs, symptom burden, and caregiving requirements may have contributed to the observed heterogeneity. Because cancer-specific profile models and cross-cancer measurement comparability were not evaluated, the same profile structure cannot be assumed to apply to each malignancy. Studies with adequate representation of individual cancer types are needed to examine this issue.

Lower monthly household income per capita was associated with lower adjusted odds of Low versus High FT profile membership, consistent with previous findings linking limited financial resources to greater FT.²¹ However, the >5000 CNY reference category included only three participants in the High FT profile and one in the Moderate FT profile. These small counts limit the precision and stability of the income contrasts. The observed direction is compatible with greater financial vulnerability among participants with lower income, but the effect magnitudes should not be treated as definitive. Previous research has associated limited financial resources with reduced access to diagnosis, treatment, and follow-up care, as well as greater perceived financial strain.²⁴ Although daytime chemotherapy may lower hospitalization-related costs,²⁵ repeated outpatient visits may be accompanied by additional expenses for transportation, accommodation, medication, and home care. These expenses may be particularly burdensome for patients with limited financial resources and may be related to greater financial vulnerability.^26,27

Participants classified as not working had lower adjusted odds of Low versus High FT profile membership than retired participants. This estimate was imprecise, with the confidence interval approaching the null value, and should be interpreted cautiously. The finding is broadly consistent with previous research on employment-related differences in FT.²⁸ Previous studies have associated stable employment with greater financial resilience, possibly reflecting the availability of a relatively stable income during treatment.²⁹ Cancer- and chemotherapy-related symptoms, including fatigue, nausea, vomiting, and peripheral neuropathy, have also been associated with reduced work capacity, sick leave, employment interruption, and productivity loss.^30,31 Under the daytime chemotherapy model, frequent outpatient treatment and management of adverse reactions may make it more difficult for some patients to maintain stable employment, particularly those without formal employment contracts or adequate social security. Such employment and income disruptions may be associated with greater financial vulnerability.³² The relatively stable pension income available to many retired patients may partly explain the observed difference between retired and nonworking patients, although this mechanism was not directly examined in the present study.⁹ Caregiving demands have also been associated with reduced working hours or labor-force withdrawal among family caregivers and lower household income.^33,34 These employment and financial circumstances may be related to greater financial distress among resource-limited families.³⁵

Disease duration was associated with FT profile membership, consistent with previous LPA findings.⁹ Compared with a disease duration of more than 1 year, shorter disease durations were associated with higher odds of membership in the Low rather than the High FT Group, suggesting that a prolonged disease course may be associated with greater financial vulnerability. However, the association was not uniform across all profile comparisons, as patients with a disease duration of less than 6 months had lower odds of membership in the Moderate rather than the High FT Group. Previous studies have associated prolonged disease courses with cumulative treatment expenditures, nonmedical costs, and income disruption, which may offer possible explanations for the observed association.^36,37 However, these potential mechanisms were not directly measured and cannot be confirmed by the present study. These findings support coordinated assistance from healthcare providers and social workers regarding treatment costs, insurance coverage, and financial assistance programs.³⁸ Strategies to address travel and time costs associated with repeated daytime chemotherapy visits warrant evaluation in future prospective studies.³⁹

Overall psychological distress differed across the profiles in unadjusted comparisons. After adjustment, higher total HADS scores were associated with greater odds of Moderate versus High FT profile membership, whereas no statistically significant association was observed for Low versus High FT membership. This pattern does not establish a monotonic relationship between psychological distress and FT severity. Previous studies have reported associations between FT and psychological distress.^40,41 However, because the measures were collected simultaneously, the observed association could reflect reciprocal relationships or shared factors, including disease burden, treatment intensity, and socioeconomic circumstances. The present study cannot determine the direction of these relationships or establish that psychosocial interventions reduce FT.

Higher Family APGAR scores were associated with greater adjusted odds of Low versus High FT profile membership, consistent with the findings of Shao et al⁴² Previous research suggests that family cohesion, communication, and shared coping may be relevant to psychological adjustment under financial stress.^43,44 However, these potential mechanisms were not directly examined in the present study. Because family functioning and FT were assessed simultaneously, better family functioning may be associated with less financial distress, while patients experiencing less financial distress may also perceive their family functioning more favorably. Both may additionally reflect shared socioeconomic or clinical circumstances. The observed association therefore does not establish that family support reduces FT. These findings support considering family functioning when assessing patients’ financial and psychosocial needs, while the effectiveness of family-oriented supportive interventions requires prospective evaluation.

Implications for Oncology Practice

The findings may inform a structured multidisciplinary pathway for identifying and responding to FT-related needs in daytime chemotherapy settings. At the initial assessment and at clinically relevant points during treatment, oncology nurses could screen for FT and review associated socioeconomic and psychosocial characteristics, including household income, employment disruption, psychological symptom burden, and family functioning. The exploratory profiles should not be used as validated clinical cutoffs or as the sole basis for allocating supportive services.

For patients reporting substantial financial concerns, nurses could initiate cost-of-care discussions, provide basic information, document identified needs, and refer them to social workers or financial navigators. Social workers could assess insurance coverage, out-of-pocket expenses, transportation and accommodation needs, employment and caregiver disruption, and eligibility for financial assistance or community resources. Patients reporting psychological distress could be referred to appropriate psychosocial services, while family discussions could be facilitated, with the patient’s consent, when limited family functioning or caregiving strain is identified. Coordination among nurses, social workers, oncologists, pharmacists, and psychosocial professionals could support individualized care planning and follow-up.

These approaches represent potential applications of the observed associations and were not evaluated as interventions in this study. Therefore, the present findings do not establish that financial counseling, psychosocial support, or family-centered care reduces FT or improves adherence or quality of life. Their feasibility, acceptability, and effectiveness should be examined in prospective implementation and intervention studies.

Limitations

Several limitations should be acknowledged. First, convenience sampling at three tertiary hospitals in Southwest China may have introduced selection bias, as participants may have differed from patients who declined or were unable to complete the questionnaires. The geographic concentration and tertiary-care setting also limit generalizability. Broader recruitment across diverse institutions and, where feasible, probability-based sampling would improve representativeness. Second, the cross-sectional design precludes conclusions about causality or temporal ordering. FT, overall psychological distress, and family functioning were assessed simultaneously, preventing evaluation of their temporal sequence and potentially reciprocal relationships. The identified profiles describe patterns at a single time point and do not establish temporal stability. Longitudinal studies, including latent transition analysis where appropriate, are needed to examine changes in profile membership. Third, self-reported measures may be affected by recall and social desirability bias. Incorporating objective financial indicators and qualitative interviews could provide a more comprehensive assessment of FT.

Fourth, several analytical choices may have influenced the findings. Treating five-point Likert indicators as continuous may have affected the profile structure, and robustness to an ordered categorical specification was not evaluated. The three-profile solution was retained for parsimony and interpretability despite better information criteria for the four- and five-profile solutions; therefore, alternative classifications remain plausible. Furthermore, although classification quality was high, the regression used most-likely profile assignment without accounting for posterior classification uncertainty, which may have affected the magnitude and precision of the estimated associations. The findings should be considered exploratory. Future analyses should assess sensitivity to indicator specification and profile number and reassess covariate associations using methods that account for classification uncertainty, such as R3STEP. Fifth, some profile-by-covariate cells, particularly those involving higher-income, retired, and self-paid categories, contained few participants. These sparse cells, together with the 28 covariate coefficients estimated across two logit equations, may have contributed to the wide confidence intervals of several aORs. Successful convergence does not ensure precise estimation, and imprecise estimates should be interpreted cautiously. In addition, no adjustment for multiple testing was applied to the category-specific regression estimates; therefore, nominal P values, particularly those close to 0.05, should be interpreted as exploratory. Larger samples with better representation across covariate categories are needed. Finally, the inclusion of multiple cancer types introduced clinical heterogeneity in prognosis, treatment, costs, symptom burden, employment consequences, and caregiving requirements. Cancer type was not included in the primary model because its additional parameters would have increased model complexity in the presence of sparse cells. Recruitment hospital was also not included, leaving potential differences in patient composition and care provision unaccounted for. Residual confounding by cancer type, hospital, and other unmeasured characteristics therefore remains possible. The pooled profiles should not be assumed to apply equally across cancer types or institutions, and their reproducibility requires further evaluation.

Conclusion

This study extends evidence on FT heterogeneity to patients with advanced cancer receiving daytime chemotherapy. Three cross-sectional FT profiles were identified, and profile membership was associated with monthly household income per capita, employment status, disease duration, overall psychological distress, and family functioning. These findings may help inform context-specific FT assessment and the development of multidisciplinary supportive approaches involving oncology nurses, social workers, and other healthcare professionals.

Funding Statement

This work was supported by the Chengdu Municipal Health Commission [Grant numbers 2025652]. The funder had no role in the study design; data collection, analysis, or interpretation; manuscript preparation; or the decision to submit the manuscript for publication.

Data Sharing Statement

The data set used in this research is available from the corresponding author upon reasonable request.

Ethical Considerations

The study was approved by the Ethics Committee of the General Hospital of the Western Theater Command of the Chinese People’s Liberation Army (Approval No. 2025EC11-ky032) and conducted in accordance with the Declaration of Helsinki.

Consent for Publication

This was deemed not applicable.

Author Contributions

Ying Zheng, Xiaoli Huang and Tao Xu contributed equally to this work and share first authorship.

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare no conflicts of interest.