Introduction
Cancer imposes a persistent emotional, logistical, and financial burden on patients’ families. When a major disaster strikes the same households, that burden multiplies: medical appointments are disrupted, caregiving networks fracture, and families must simultaneously manage recovery and illness.Reference McCarthy, Pyke-Grimm and Feraco1–Reference Francois2
Maintaining or improving family well-being is a core objective in cancer care and becomes even more critical in disaster settings. Family well-being influences mental health outcomes, treatment adherence, caregiving capacity, and overall quality of life for both patients and caregivers.Reference Ferrell, Kravitz and Borneman3 When families face compounded stress from illness and external shocks, protecting well-being may help sustain resilience and long-term recovery.Reference Wada, Takebayashi and Murakami4
Japan offers a unique context for studying this dual burden—that is, the simultaneous strain of family cancer care and recurrent disasters. Since the 2011 Great East Japan Earthquake and the Fukushima Daiichi nuclear accident, which is a typical natural hazard-induced technological (natech) event, the country has faced repeated typhoons, floods, and earthquakes, with lasting psychological effects (i.e., the long-term psychological burden of disasters, including persistent symptoms of depression, anxiety, and post-traumatic stress).Reference Yamashita and Shigemura5–Reference Kunii, Usukura and Utsumi7
Despite increasing recognition that disaster exposure impacts mental health, little work has examined how post-disaster recovery situations (unexposed, recovered, unrecovered) interact with cancer-related family stressors to shape well-being. Conservation of Resources (COR) theory, proposed by Hobfoll, posits that psychological stress arises when individuals lose, fail to gain, or are threatened with the loss of valued resources such as housing, income, social ties, and health.Reference Hobfoll8–Reference Hobfoll, Halbesleben and Neveu9 In this study, subjective disaster recovery and satisfaction with cancer care were conceptualized as key material and psychosocial resources that may shape family well-being. Drawing on this framework and prior empirical findings, the following hypotheses were formulated:
-
• H1: Unrecovered disaster damage is associated with lower overall well-being; families with recovered or no damage report higher well-being.
-
• H2: Greater satisfaction with cancer care (Family Satisfaction with Advanced Cancer Care [Short FAMCARE Scale]) is positively associated with overall well-being.
-
• H3: The positive impact of high FAMCARE is expected to be strongest in households whose recovery from disaster damage has been achieved or is in prospect and comparatively weaker in those that see no prospect of recovery (consistent with COR theory and prior disaster-health research).
Because well-being is multidimensional and disaster recovery is plausibly correlated with several covariates, this study used confirmatory factor analysis to establish a measurement model and structural equation modelling to test the joint effects of disaster recovery and cancer care on the resulting latent well-being construct. Measurement invariance was assessed first to ensure that any cancer-vs-non-cancer comparison is interpretable rather than an artifact of differential item functioning. Residency in Fukushima Prefecture was used as a strategic recruitment criterion to reach households likely to have experienced prolonged disaster-related disruption; the aim was not to estimate the specific impact of the 2011 Great East Japan Earthquake but to examine recovery from any major disaster of the preceding 10 years. By clarifying when and for whom satisfaction with cancer care is accompanied by better well-being, this study offers guidance for oncology support programs and post-disaster family services.
Methods
Study Design, Setting, and Framework
This study employed a cross-sectional web survey fielded in September 2019. The exposure window covered any major disaster experienced during the preceding 10 years, so the 2011 Great East Japan Earthquake and the Fukushima Daiichi Nuclear Power Station accident are the earliest and most widely experienced events referenced by respondents, but not the only ones. Fukushima Prefecture was used as a recruitment stratum because the 2011 events produced widespread destruction, long-term evacuation, and continuing social and health challenges there.
The model was guided by COR theory, which views well-being as shaped by the loss and recovery of personal, material, and relational resources. Disaster recovery and satisfaction with cancer care were conceptualized as key resources. Structural equation modeling (SEM) was used to test associations and interactions among these factors.
Participants and Power Justification
Adults aged 20 to 69 years residing in Japan were recruited nationwide through a commercial online research panel operated by Intage Research Inc. between 26 and 29 September 2019. The survey was designed to sample respondents from across Japan rather than from a single affected region. To ensure adequate subgroup representation, quotas were set for 4 approximately equal cells defined by Fukushima residency (yes/no) and household cancer experience (yes/no). Fukushima residency was included as a contextual stratification factor because it represents a setting with well-documented long-term disaster-related disruption; however, the study itself was not restricted to Fukushima and aimed to examine disaster recovery more broadly across Japan. This quota-based approach was adopted to improve subgroup coverage rather than to achieve national representativeness. Recruitment continued until the overall target was reached, but the 4 cells did not fill equally. Against a target of approximately 545 per cell, the realized cells were 575 respondents outside Fukushima without family cancer, 749 in Fukushima without family cancer, 527 outside Fukushima with family cancer, and 328 in Fukushima with family cancer (Table 1). Recruitment in cells that were still filling was therefore not halted once their own target was met, and the Fukushima and family-cancer cell never reached its target, so the realized design is unbalanced with respect to the joint distribution of Fukushima residency and family cancer experience. After quality control procedures, the research company delivered a final dataset of 2,179 individuals who completed the web survey and provided electronic informed consent. The protocol was approved by the ethics committees of Minamisoma Municipal General Hospital (approval #5-14) and Fukushima Medical University (approval #2019-198), and adhered to the principles of the Declaration of Helsinki.
Participant characteristics by family cancer experience†
Table 1. Long description
The table includes columns for Characteristic, Total (N = 2,179), No family cancer (N = 1,324), Family cancer (N = 855), Test statistic, df, and P-value.
Key findings include:
* Age: Mean age is approximately 46.7 for both groups (P = 0.836).
* Household income: Significant difference (P = 0.001) with 37.8% of the family cancer group earning 6 million JPY or above compared to 30.7% in the no family cancer group.
* Fukushima residency: Significant difference (P < 0.001) where 61.6% of the family cancer group lives outside Fukushima compared to 43.4% of the no family cancer group.
* Sex: No significant difference (P = 0.892); approximately 55% male in both groups.
* K 6 psychological distress: Mean scores are 6.4 for no family cancer and 6.6 for family cancer (P = 0.511).
* Subjective recovery status: No significant difference (P = 0.595); most participants are unexposed (78.6% total).
* Subjective health: No significant difference (P = 0.897); approximately 63% report low subjective health.
* Education: Significant difference (P < 0.001) with 67.1% of the family cancer group having higher education compared to 58.4% in the no family cancer group.
* Comorbidity: Significant difference (P < 0.001) with 51.3% of the family cancer group reporting comorbidities compared to 42.8% in the no family cancer group.
† Fukushima residency reflects respondents’ residence at the time of the 2019 survey. Respondents who had previously resided in Fukushima at the time of the 2011 disaster but had moved outside Fukushima by 2019 are classified under “Outside Fukushima.” Categorical variables were compared using chi-square tests; continuous variables were compared using Welch t-tests. Subjective recovery status was recorded on a 4-level item and dichotomized: “Recovered” combines respondents who had almost fully recovered with those who had not yet recovered but could see recovery ahead, and “unrecovered” combines those who saw no prospect of recovery with those who considered recovery impossible. Household income was supplied by the panel provider in 14 ordered bands of unequal width; it is shown here in 3 collapsed categories and was entered in the models as the standardized band score.
Household cancer experience was ascertained from the survey item asking whether any family member had been diagnosed with cancer during the preceding 10 years, irrespective of co-residence; respondents indicated the relationship of the diagnosed member (parent, child, grandparent, grandchild, sibling, or spouse) or that no family member had been diagnosed, and respondents who declined to answer did not proceed. Prefecture of residence and household income were not asked in the questionnaire; both were supplied by the research panel provider as registered attributes, income in fourteen ordered bands. For brevity, “cancer-affected household” is used throughout to denote a respondent reporting a family member diagnosed with cancer in the preceding ten years, whether or not that member lived with the respondent.
Respondents who had been internally displaced from Fukushima after the 2011 disaster were classified by their current residence at the time of the 2019 survey; thus, evacuees living outside Fukushima at survey time were classified as “outside Fukushima.”
An a priori power analysis was conducted for a multiple-regression model with 21 predictors, including interaction terms. Assuming a small effect size (f2 = 0.02), a two-tailed α of 0.05, and 90% power, the required sample size was 1,347. The final analytic sample of 2,179 respondents exceeded this requirement and was therefore considered sufficient for the planned analyses. This calculation concerns the power of the full sample to detect a small effect in the overall model and does not extend to the subgroup interaction between satisfaction with cancer care and subjective recovery status, which is estimated from small cells (see Limitations) and is correspondingly underpowered.
Analytical Data
Dependent variables
Well-being. Well-being was measured using 17 questionnaire items adapted from validated Japanese versions of the Gallup World Poll and the European Social Survey (ESS) modules. Following theoretical frameworks that distinguish between hedonic and eudaimonic components of well-being,Reference Ryff10–Reference Nettle12 the items were organized into 3 conceptual domains: (1) momentary emotional states (6 items, including enjoyment, emotional happiness, laughter, and reverse-coded stress, sadness, worry); (2) life evaluation (2 items: life satisfaction and general happiness, based on Cantril’s ladder)Reference Cantril13; and (3) psychological resources (9 items from the ESS, including vitality, resilience, emotional stability, optimism, self-esteem, engagement, meaning, positive relationships, and competence).14 Because only the respondent completed the survey, well-being throughout this study is measured at the level of the responding individual and is used as an indicator of well-being within a cancer-affected or non-cancer household, rather than as a measure of the household as a unit or of the diagnosed family member’s own well-being.
A 2-factor structure was adopted, on theoretical grounds and after examining alternative structures (Supplementary Table S1). A negative-free emotion factor captured the absence of recent distress (sadness, worry), while a broader resource-oriented well-being factor encompassed life evaluation and psychological resources—a separation consistent with prior between-country comparisons showing that negative affective states form a distinct cluster from positive emotion and psychological resources.Reference Machida, Murakami and Kashiwazaki15 Accordingly, well-being was treated as a higher-order latent construct composed of these 2 first-order factors, and this composite served as the primary outcome in all structural models. The measurement model used in all structural analyses fitted the data acceptably (robust CFI 0.952, robust RMSEA 0.044, SRMR 0.026; the configural model in Supplementary Table S2), and the reliability, convergent and discriminant validity, and measurement invariance evidence reported below all support it. The retained items differ in response format, from binary items to eleven-point scales, so item means in Table 2 and unstandardized loadings in Supplementary Table S3 are not comparable in magnitude across items; standardized loadings are reported alongside them for that reason.
Descriptive statistics and reliability of well-being items by family cancer experience†
Table 2. Long description
The table presents descriptive statistics and reliability for two main scales.
1. Negative-free emotion scale (alpha 0.650, omega 0.650, mean inter-item r 0.480, Spearman-Brown lambda 0.650):
- The absence of sadness: No family cancer (Mean 0.714, S D 0.452) vs. Family cancer (Mean 0.673, S D 0.470). t equals 2.030, P-value 0.042, Cohen’s d -0.089.
- The absence of worry: No family cancer (Mean 0.477, S D 0.500) vs. Family cancer (Mean 0.426, S D 0.495). t equals 2.330, P-value 0.020, Cohen’s d -0.102.
2. Resource-oriented well-being scale (alpha 0.780, omega 0.840):
- Life satisfaction: No family cancer (Mean 5.880, S D 2.440) vs. Family cancer (Mean 5.910, S D 2.420). t equals -0.288, P-value 0.773, Cohen’s d 0.013.
- General happiness: No family cancer (Mean 6.170, S D 2.510) vs. Family cancer (Mean 6.210, S D 2.500). t equals -0.418, P-value 0.676, Cohen’s d 0.018.
- Emotional stability: No family cancer (Mean 2.410, S D 0.806) vs. Family cancer (Mean 2.390, S D 0.791). t equals 0.555, P-value 0.579, Cohen’s d -0.024.
- Vitality: No family cancer (Mean 2.120, S D 0.752) vs. Family cancer (Mean 2.150, S D 0.721). t equals -0.955, P-value 0.340, Cohen’s d 0.042.
- Competence: No family cancer (Mean 2.840, S D 0.954) vs. Family cancer (Mean 2.890, S D 0.950). t equals -1.370, P-value 0.172, Cohen’s d 0.060.
† Despite reaching nominal significance, the absolute magnitude of the negative-free emotion difference between groups is small (Cohen’s d < 0.15). Items differ in response format, so means are not comparable in magnitude across items. Possible ranges: absence of sadness and absence of worry 0-1; life satisfaction and general happiness 0-10; emotional stability and vitality 1-4; competence 1-5.
Exposure factors
Disaster recovery status. Subjective recovery status was derived from 3 consecutive survey items. Respondents were first asked whether they, or anyone close to them, had experienced a major disaster during the preceding 10 years. A major disaster was defined for respondents as an event causing 100 or more deaths or missing persons, with the Great East Japan Earthquake and the Kumamoto Earthquake given as Japanese examples; floods, landslides, and typhoons therefore qualified only when they reached that scale. Five response options distinguished no such experience; a major disaster nearby that did not affect the respondent; damage to the respondent’s or their family’s house or farmland; injury or illness of the respondent or a family member; and death of a family member or of someone important to them. Respondents in the latter 3 categories were then asked how far they had recovered from that damage, choosing among “almost fully recovered,” “not yet recovered but recovery is in sight,” “no prospect of recovery,” and “recovery will not happen, and there is no choice but to accept it.” For analysis, respondents reporting no experience or an unaffecting nearby event were classified as unexposed (n = 1,712). The remaining respondents were dichotomized according to whether recovery had been achieved or was foreseeable: those answering “almost fully recovered” (n = 306) or “not yet recovered but recovery is in sight” (n = 66) were classified as recovered (n = 372), and those answering “no prospect of recovery” (n = 35) or “recovery will not happen and there is no choice but to accept it” (n = 60) as unrecovered (n = 95). This cut separates households that can foresee the restoration of what they lost from those that cannot, which is the distinction COR theory makes central, and it avoids modelling the 2 smallest categories separately. Throughout this paper, therefore, “recovered” denotes recovery that has been achieved or is in prospect rather than a completed return to pre-disaster conditions, and “unrecovered” denotes damage for which the respondent saw no prospect of recovery. Respondents reporting any disaster experience were additionally asked to identify the event or events involved, choosing among the Great East Japan Earthquake (2011), the Kumamoto Earthquake (2016), the July 2018 heavy rain and flooding, or another event specified in free text; multiple selections were permitted. Among the 467 respondents classified as exposed, 409 identified the Great East Japan Earthquake, 31 the July 2018 heavy rain and flooding, 11 the Kumamoto Earthquake, and 27 another event, with 11 respondents naming 2 events. Flooding therefore entered the sample where it reached the scale specified in the exposure item, as the July 2018 event did, whereas the smaller and more frequent floods and landslides that occur in Japan did not qualify. The survey was fielded on September 26-29, 2019 and therefore predates Typhoon Hagibis (the 2019 East Japan Typhoon), which struck Japan on 12 October 2019 and caused severe flooding in Fukushima among other prefectures; that event postdates the survey and is not among the disasters to which respondents could have referred.Reference Yoshida, Ozaki and Morita16 Cancer experience was coded dichotomously from the family-cancer item described above. Among the 189 respondents with a family cancer history who were classified as exposed, the relative timing of diagnosis and disaster could be established for 159: the diagnosis preceded the disaster for 49, fell in the same year for 15, and followed it for 95. Timing could not be established for the remaining 30, comprising 16 with an unknown year of diagnosis and 14 who identified only a free-text event.
When multiple family members had cancer, respondents were instructed to answer FAMCARE items with reference to the most recently diagnosed family member, regardless of co-residence. The survey did not separately ascertain whether the respondent was the primary caregiver or how directly they were involved in care; therefore, FAMCARE responses should be interpreted as respondent-level family perceptions of the cancer care received by the affected family member. Satisfaction with cancer care was measured using a Japanese translation of the 10-item short-form Family Satisfaction with Advanced Cancer Care Scale (Short FAMCARE Scale).Reference Kristjanson17–Reference Ornstein, Teresi and Ocepek-Welikson18 A Ramsey RESET test (F = 8.15, df = 2/838, P < 0.001) and Johnson-Neyman plots (Supplementary Figure S1) indicated nonlinearity in its association with well-being. To accommodate this nonlinearity, FAMCARE scores were divided into three categories based on the 33rd and 66th percentiles (i.e., tertiles), forming low, mid, and high satisfaction groups. Because the score distribution is discrete and heavily concentrated, with 340 of 855 respondents scoring exactly 10, the groups were not equal in size: 297 respondents scored 8 or below (low), 413 scored 9 or 10 (mid), and 145 scored 11 or above (high). Cancer stage was reported by 520 of the 855 respondents with a family cancer history (60.8%): 50 in situ or intraepithelial, 130 stage I, 105 stage II, 87 stage III, and 148 stage IV. A further 305 respondents (35.7%) answered that they did not know the stage and 30 (3.5%) declined to state it. The year of diagnosis was reported by 716 respondents (83.7%). The family member referred to was living at the time of the survey for 445 respondents and had died for 396, of whom 307 had died of cancer; 14 respondents declined to answer. FAMCARE responses therefore combine appraisals of ongoing and of completed care.
All analyses adjusted for age (standardized), sex, education (junior high or high school; vocational school, junior college, technical college, university or graduate school; or prefer not to answer), occupation (company employees, civil servants, teachers and other professionals; self-employed respondents, including those in farming, forestry and fishing; or others, comprising part-time workers, homemakers, students, and unemployed or retired respondents), marital status (never married; married, including respondents living apart from their spouse; or other, comprising divorced and widowed respondents), household income (supplied by the panel provider in 14 ordered bands and treated as continuous after standardization), Fukushima residency, subjective health (high, comprising “very good” and “good,” vs. low, comprising “fair,” “poor,” and “very poor”), the 6-item Kessler Psychological Distress Scale (K6; continuous), and comorbidity (any of 10 chronic conditions vs. none).Reference Marmot19–Reference Rijken, van Kerkhof and Dekker24 Subjective health, K6, and comorbidity describe the respondent, not the family member diagnosed with cancer. Seventy respondents who selected only “prefer not to answer” for the comorbidity item were included in the comorbidity-present category.
Data Analyses
Two model-based methods underpin the analyses below. Categorical factor analysis is a factor analysis for ordinal (categorical) item responses that identifies the common latent factors underlying the observed items; here it was used to determine how many well-being factors best represent the data and which items load on each factor. Covariance structure analysis (structural equation modelling) simultaneously models the relationships among observed and latent variables, allowing estimation of not only direct effects but also indirect effects operating through mediating variables; here it was used to test how subjective disaster recovery and satisfaction with cancer care jointly relate to latent well-being.
Measurement Model: Categorical Factor Analysis, Reliability, and Invariance Testing
All analyses were run in R 4.4.1 (lavaan, semTools, psych, car, interactions) under robust maximum-likelihood estimation (MLR) with 2-tailed α = 0.05; weighted least squares mean- and variance-adjusted (WLSMV) estimation gave virtually identical results (Supplementary Table S1).Reference Li25 Continuous predictors were z-standardized and categorical ones dummy-coded.
The initial pool of 17 well-being items was screened by an initial CFA screen: 7 items met the loading threshold (λ ≥ 0.60) and were retained,Reference Hair, Black and Babin26 whereas the other 10 were discarded. For each retained item, the mean and standard deviation within cancer-affected and non-affected households were calculated, and group means were compared using Welch’s t tests.
Internal consistency for each first-order factor was considered acceptable when Cronbach’s alpha and McDonald’s omega were >= 0.70.Reference Hayes and Coutts27 Because the negative-free emotion factor contains only 2 items, reliability was assessed instead with the mean inter-item correlation and Spearman-Brown lambda, which outperform alpha/omega for 2-item scales.Reference Eisinga, te Grotenhuis and Pelzer28
Convergent validity was evaluated using standardized loadings, composite reliability, and Average Variance Extracted (AVE),Reference Lee29 and discriminant validity using the Fornell-Larcker criterion and the Heterotrait-Monotrait Ratio (HTMT).Reference Voorhees, Brady and Calantone30 For the 2-item negative-free emotion factor, reliability was assessed primarily with Spearman-Brown lambda. Justifiable residual covariances among conceptually overlapping items were retained to improve fit without violating theory.Reference Pan, Ip and Dubé31 Measurement invariance across cancer-affected and non-affected households was tested sequentially (configural → metric → scalar → strict) with ΔCFI < 0.01 as the criterion.Reference Dimitrov32 Sensitivity analyses using a bifactor model and a 5-factor alternative supported the stability of the 2-factor solution (Supplementary Table S1).Reference Khojasteh and Lo33
Structural Models and Moderation Analysis
A multigroup SEM was fitted to examine whether the association between subjective recovery status and well-being varied by household cancer status. The second-order well-being factor was regressed on subjective recovery categories and the full set of covariates described earlier. All structural paths were freely estimated by group, using the validated measurement model with residual covariances retained as specified in the measurement-invariance step described above.
A second model focused on cancer-affected households, incorporating FAMCARE (tertile-coded) and its interaction with subjective recovery status to test moderation. For both models, standardized regression coefficients, 95% confidence intervals, and two-tailed p-values are reported for all structural predictors. For both the multigroup SEM and the single-group interaction model, latent R2 for well-being was extracted, and Cohen’s f2 was calculated for each key path; following Cohen’s (1988) benchmarks, f2 values of 0.02, 0.15, and 0.35 were interpreted as small, medium, and large effects.Reference Cohen34 Multicollinearity among exogenous covariates was assessed via variance inflation factors (VIFs) to ensure no concerning collinearity. The analysis tested the 3 pre-specified hypotheses (H1-H3) without formal adjustment for multiple comparisons; because the interaction and contrast P-values (0.04 and 0.014) lie close to the 0.05 threshold and depend on small cells, they are interpreted as exploratory rather than confirmatory.
Results
Participant characteristics are shown in Table 1. The final analytic sample included 2,179 respondents, of whom 855 reported a family cancer history and 1,324 did not. The family-cancer group differed from the non-cancer group in Fukushima residency, household income, educational attainment, and comorbidity, whereas age, sex, subjective recovery status, subjective health, occupation, marital status, and K6 psychological distress did not differ meaningfully between groups.
As shown in Table 2, families with cancer-affected members reported significantly lower levels of negative-free emotion than those without, specifically for absence of sadness (t = 2.03, P = 0.042) and absence of worry (t = 2.33, P = 0.020). Despite nominal significance, these differences were small in magnitude (Cohen’s d < 0.15). In contrast, none of the resource-oriented well-being items showed statistically significant differences. These small negative-free emotion differences are visualized in Supplementary Figure S2.
In plain terms, both well-being factors were measured reliably. Confirmatory factor analysis showed that all standardized loadings exceeded 0.63, with composite reliability of 0.81 to 0.82 for the resource-oriented factor and 0.62 to 0.66 for the 2-item negative-free emotion factor, for which the Spearman-Brown lambda of 0.65 is the more appropriate index; convergent validity at the second-order well-being level was adequate (Supplementary Table S3). Discriminant validity was supported by both the Fornell–Larcker criterion and HTMT estimates: the square root of AVE for each construct exceeded the inter-factor correlation, and the HTMT value was below the recommended threshold of 0.85 (Supplementary Table S4). Measurement invariance was supported sequentially through configural, metric, scalar, and strict levels (all ΔCFI < 0.01, ΔRMSEA ≤ 0.002, ΔSRMR ≤ 0.001), indicating that the well-being questionnaire behaved similarly in cancer-affected and non-cancer households (Supplementary Table S2). Group comparisons of well-being therefore reflect substantive differences rather than measurement artefacts.
Figure 1 summarizes the focal structural paths, covariate adjustment, and main numerical findings from the multigroup and cancer-affected household models. In Panel A, unrecovered subjective recovery status was associated with lower well-being only among non-cancer households (β = −0.08, P = 0.001), whereas Panel B shows that high satisfaction with cancer care was associated with higher well-being in cancer-affected households (β = 0.17, P < 0.001), with an additional benefit among recovered households (High × Recovered: β = 0.09, P = 0.04; direct contrast estimate = 0.149, 95% CI 0.031-0.268, P = 0.014; the corresponding model-predicted cell means are shown in Supplementary Figure S3).
Summary of the structural equation models and focal results. Panel A shows the multigroup model comparing households with and without family cancer history. Panel B shows the cancer-affected household model including the Short FAMCARE × subjective recovery interaction. Solid arrows indicate focal structural paths; dashed arrows indicate covariate adjustment. Values on structural arrows indicate standardized coefficients; the direct contrast is shown in model units. In both panels, “recovered” denotes recovery that has been achieved or is in prospect, and “unrecovered” denotes damage for which the respondent saw no prospect of recovery.
Figure 1 Long description
Figure 1 is a two-panel conceptual path diagram summarising the structural equation models and focal results. Panel A (multigroup model, estimated separately for households with and without a family cancer history) shows a box, subjective recovery status (unexposed, recovered, or unrecovered), with an arrow pointing to a box, overall family well-being. An annotation states that unrecovered disaster damage was associated with lower well-being only in non-cancer households (beta = -0.08, p = 0.001). A measurement-structure box indicates that overall family well-being comprises a negative-free emotion factor (absence of sadness, absence of worry) and a resource-oriented well-being factor (life satisfaction, general happiness, emotional stability, vitality, competence). A covariates box lists the adjustment set: age, sex, income, K6 distress, subjective health, comorbidity, Fukushima residence, education, occupation, marital status. Panel B (cancer-affected households) shows Short FAMCARE satisfaction (low, middle, high) and subjective recovery status pointing, through a FAMCARE-by-subjective-recovery interaction term, to overall family well-being. Annotations report the Short FAMCARE main effect (high versus low, beta = 0.17, p < 0.001) and the high-by-recovered interaction (beta = 0.09, p = 0.04). A direct-contrast box reports high satisfaction with recovery versus high satisfaction without recovery, contrast estimate 0.149 (95% CI 0.031 to 0.268, p = 0.014). The same covariate set as Panel A is applied.
Structural estimates for overall well-being are shown in Table 3 for households with and without a family cancer history. Across both groups, higher K6 scores were consistently associated with lower well-being (β = −0.77 to −0.81, P < 0.001), whereas better subjective health was positively associated with well-being (β = 0.23-0.25, P < 0.001). Among non-cancer households, unrecovered disaster exposure was significantly associated with lower well-being (β = −0.08, 95% CI −0.13 to −0.03, P = 0.001; Cohen’s f2 = 0.06). This association was not observed in cancer-affected households (β = 0.00, 95% CI −0.07 to 0.07, P = 0.98; Cohen’s f2 < 0.01). Household income was positively associated with well-being in both groups, with a stronger association among cancer-affected households.
Standardized structural regression estimates for overall well-being in households with and without a family cancer experience†
Table 3. Long description
The table presents standardized and unstandardized regression estimates for overall well-being across two groups: No family cancer and Family cancer. Columns for each group include Std.Est, Unstd, 95% C I (Unstd), P-value, and V I F.
Key predictors and findings include:
* Age: No significant effect in either group (P-values 0.91 and 0.34).
* Income: Positive correlation in both groups, stronger in Family cancer (Std.Est 0.11, P = 0.001) than No family cancer (Std.Est 0.05, P = 0.05).
* K 6 (Psychological Distress): Strong negative correlation in both groups (Std.Est -0.77 and -0.81, both P < 0.001).
* Subjective health: High health significantly predicts higher well-being in both groups (P < 0.001).
* Comorbidity: Significant positive predictor only for the Family cancer group (Std.Est 0.11, P = 0.001).
* Fukushima residency: Living in Fukushima is a significant negative predictor only for the Family cancer group (Std.Est -0.08, P = 0.01).
* Subjective recovery status: Being unrecovered is a significant negative predictor for the No family cancer group (Std.Est -0.08, P = 0.001) but not for the Family cancer group (P = 0.98).
* Education: Higher education is significant only for the No family cancer group (Std.Est 0.06, P = 0.02).
* Marital status: Being married is a significant positive predictor only for the No family cancer group (Std.Est 0.13, P < 0.001).
* Occupation: No significant differences found relative to self-employed reference in either group.
† Std.Est., Standardized regression coefficient (β); higher values indicate higher well-being. Unstd, unstandardized coefficient (in well-being units); CI, confidence interval for the unstandardized coefficient; VIF, the generalized variance inflation factor raised to the power 1/(2·Df), which for single-degree-of-freedom terms equals the square root of the variance inflation factor and is on the scale of a standard error inflation factor. “Ref.” indicates the reference category (coded 0) against which other categories are compared. P-values are reported within each group because the model was estimated separately for cancer-affected and non-cancer households. Cohen’s f2 for the effect of unrecovered disaster exposure on overall well-being was 0.06 in non-cancer households and <0.01 in cancer-affected households.
Practically, families who saw no prospect of recovery from their disaster damage reported about a tenth of a standard deviation lower well-being than unexposed families, representing a small but consistent gap that was still present at the time of the 2019 survey, up to 10 years after the disaster to which respondents referred.
Table 4 presents standardized regression estimates from the structural equation model fitted to cancer-affected households. This model included the Short FAMCARE Scale, subjective recovery status, and their interaction. Greater satisfaction with cancer care was associated with higher well-being: relative to the low-satisfaction reference group, the middle category showed a modest gain in well-being (β = 0.10; unstandardized 0.05, 95% CI 0.01 to 0.08, P = 0.005), and the high category showed a larger gain (β = 0.17; unstandardized 0.10, 95% CI 0.05 to 0.15, P < 0.001), supporting H2.
Standardized regression estimates for overall well-being in cancer-affected households
Table 4. Long description
The table contains six columns: Predictor, Std.Est, Unstd, 95% C I, P-value, and V I F.
Key findings include:
* Age: Std.Est 0.00, P-value 0.91, V I F 1.24.
* Income: Std.Est 0.10, P-value 0.003, V I F 1.13.
* K 6: Std.Est -0.79, P-value < 0.001, V I F 1.14.
* Subjective health: High health has a Std.Est of 0.23 and P-value < 0.001 compared to the Low reference.
* Comorbidity: With comorbidity has a Std.Est of 0.11 and P-value 0.001 compared to the Without reference.
* Sex: Female has a Std.Est of 0.01 and P-value 0.83 compared to Male reference.
* Fukushima residency: Fukushima residency has a Std.Est of -0.08 and P-value 0.02 compared to Outside Fukushima reference.
* Subjective recovery status: Recovered (Std.Est 0.00, P 0.95) and Unrecovered (Std.Est 0.02, P 0.68) compared to Unexposed reference.
* Education: Vocational/University (Std.Est -0.02, P 0.49) and Prefer not to answer (Std.Est -0.02, P 0.38) compared to Junior/High school reference.
* Marital status: Married (Std.Est 0.05, P 0.20) and Other (Std.Est -0.05, P 0.10) compared to Never married reference.
* Occupation: Company employee (Std.Est -0.02, P 0.75) and Other (Std.Est 0.08, P 0.28) compared to Self-employed reference.
* Short FAMCARE Scale: Middle (Std.Est 0.10, P 0.005) and High (Std.Est 0.17, P < 0.001) compared to Low reference.
* Interaction terms: The interaction between High FAMCARE and Recovered status shows a Std.Est of 0.09 and P-value 0.04.
† Cohen’s f2 for the effect of the interaction between high satisfaction with cancer care and recovered subjective recovery status on overall well-being in cancer-affected households was 0.07. Std.Est, Standardized estimate; Unstd, unstandardized estimate; CI, confidence interval for unstandardized estimate; P, P-value; VIF, the generalized variance inflation factor raised to the power 1/(2·Df), which for single-degree-of-freedom terms equals the square root of the variance inflation factor and is on the scale of a standard error inflation factor.
The size of this benefit depended on subjective recovery status. The interaction term High × Recovered was positive and significant (β = 0.09; unstandardized 0.12, 95% CI 0.01 to 0.22, P = 0.04; Cohen’s f2 = 0.07), indicating that families who reported recovery as achieved or in prospect derived an additional well-being benefit from high satisfaction with cancer care. No comparable moderation was observed in the unrecovered stratum (Mid × Unrecovered: β = −0.01; unstandardized −0.03, 95% CI −0.17 to 0.11, P = 0.67; High × Unrecovered: β = −0.02; unstandardized −0.05, 95% CI −0.20 to 0.09, P = 0.46).
A direct contrast confirmed this moderation: the well-being advantage of high satisfaction with cancer care was significantly greater in households recovered from disaster damage than in those unrecovered (contrast estimate = 0.149, 95% CI 0.031-0.268, P = 0.014; Supplementary Table S5). Sensitivity analysis using Low × Recovered as the reference category yielded the same substantive conclusion (Supplementary Table S6). The corresponding model-predicted cell means are shown in Supplementary Figure S3.
Put simply, high satisfaction with cancer care was associated with better family well-being, but only meaningfully so in households whose recovery from disaster damage had been achieved or was in prospect.
Discussion
Using a large, quota-balanced sample of adults living inside and outside Fukushima, this study examined how subjective recovery status and family cancer experiences jointly shape well-being. The central finding is that the two stressors interact rather than add: satisfaction with cancer care translates into measurable well-being gains only in households that reported recovery from the disaster as achieved or in prospect. Recovery is subjective throughout this study: respondents judged their own recovery, and the recovered category combines those who had almost fully recovered with those who had not yet recovered but could see recovery ahead. In cancer-affected households, well-being rose monotonically across the low, middle, and high Short FAMCARE tertiles, and the High × Recovered cell carried an additional gain (contrast estimate = 0.149, 95% CI 0.031-0.268, P = 0.014; Cohen’s f2 = 0.07) over and above the FAMCARE main effect.
The contrasting pattern in non-cancer households helps clarify what this interaction means. There, unrecovered status itself predicted a small-to-moderate well-being decline at the time of the survey, up to 10 years after the disaster to which respondents referred, whereas respondents who had escaped major damage, or whose recovery was achieved or in prospect, did not differ from one another (recovered vs. unexposed among non-cancer households: β = 0.02, P = 0.51; Table 3). In cancer-affected households, by contrast, the chronic burden of cancer caregiving appears to dominate, masking the marginal effect of disaster recovery unless care quality is high. In COR terms, satisfaction with cancer care functions as a psychosocial resource that can only be converted into well-being once the household can regard the more fundamental material resources of housing and livelihood as restored, or as securely on the way to being restored.Reference Hobfoll8 Consistent with this reading, household income was independently associated with well-being in cancer-affected households (β = 0.11, P = 0.001), suggesting that financial resources contribute beyond perceived care quality even though the Short FAMCARE Scale does not directly assess financial strain.Reference Ozaki, Murakami and Nomura35
The pattern shown in Supplementary Figure S3 deserves separate comment. Model-predicted well-being was highest in the cell combining high satisfaction with cancer care and reported recovery, exceeding not only the unrecovered cells but also the corresponding unexposed cell. Two readings are possible. First, households that have been through both a disaster and a cancer diagnosis, and that can see their recovery as achieved or within reach, may judge their present situation against a more demanding point of comparison, so that similar circumstances are reported as greater well-being; such reappraisal, described in the literature as benefit finding or as response shift, is documented among survivors of disasters and of serious illness alike, and in COR terms corresponds to a gain spiral following successful restoration of lost resources. Second, the estimate rests on small cells. Among the 855 cancer-affected households, high satisfaction was reported by 112 unexposed, 28 recovered, and 5 unrecovered respondents, so the contrast that defines the additional benefit compares 28 households with 5. The interaction term and the direct contrast both reached conventional significance, but with cells of this size the magnitude is imprecisely estimated and sensitive to individual observations. Cross-sectional data cannot separate the two readings, and the additional benefit should be treated as a hypothesis for prospective study rather than as an established effect.
Clinical implications
These findings have direct implications for post-disaster recovery planning. Because unresolved subjective recovery and lower household income were each independently linked to lower well-being, local recovery policies should prioritize tangible drivers of subjective recovery—such as completing housing repair and stabilizing employment—which the broader disaster-recovery literature identifies as the dominant contributors to a sense of restored normalcy. Cancer-affected households may warrant explicit attention within such planning: the well-being benefits of high-quality cancer care were observed only where recovery was achieved or at least in prospect, a pattern consistent with co-delivering reconstruction support and oncology care continuity rather than treating them as sequential, though the cross-sectional design precludes causal inference.Reference Ozaki, Leppold and Tsubokura36–Reference Kobayashi, Ozaki and Kotera38
Limitations
This study is cross-sectional, so causal directions cannot be inferred, and all measures were self-reported, with possible recall bias for disaster losses reported up to 10 years after the event. In addition, the use of an online research company’s registered panel introduces possible selection bias, as panel members may differ from the general population in digital access or engagement, and individuals aged 70 and older were excluded, limiting generalizability to older populations. The realized quota cells were also unbalanced: cancer-affected households were substantially under-represented among Fukushima residents (328 of 1,077), so any estimate involving the joint distribution of Fukushima residency and family cancer experience rests on unequal cell sizes and should not be read as representative of cancer-affected households in Fukushima.
The year of diagnosis and the disasters referred to were both recorded, so the relative timing of diagnosis and disaster is described above rather than assumed. It was not used to stratify the moderation analysis: cross-classifying timing by subjective recovery status and satisfaction tertile leaves 7 of 18 cells with fewer than 5 respondents and 3 cells empty. The cross-sectional estimates should therefore be read as average associations across families whose cancer preceded, coincided with, or followed the disaster. The onset year is known for the three named events but not for the 27 respondents who described another event in free text, so the time elapsed since exposure varies within the exposed group and heterogeneity in it is absorbed into the 3-level recovery variable. Eleven respondents named 2 events, for whom the recovery item cannot be attributed to a single disaster. Subjective recovery was recorded on a 4-level item and analyzed as a binary contrast between recovery that had been achieved or was foreseeable and recovery that was not foreseeable; a finer gradation, and in particular the distinction between households that had already recovered and those that expected to recover, was not modeled. Respondents internally displaced from Fukushima after 2011 were classified by their 2019 residence, and the survey did not separately enumerate evacuees living outside Fukushima at survey time, which may obscure their distinct exposure history.
Cancer stage was recorded, but 335 of the 855 respondents with a family cancer history (39.2%) either did not know it or declined to state it, and among those who did report it, only 148 described stage IV disease. The Short FAMCARE Scale was developed for advanced cancer care, and the short form in particular was derived from an item bank based on family members of patients with advanced cancer, whereas this sample spans the full range of severity and includes families whose relative was still in treatment; the satisfaction ratings should therefore be read as family perceptions of cancer care in general rather than of advanced or end-of-life care specifically. Stage was therefore not included as a covariate: adjusting for it would restrict the cancer-affected subsample to the 520 respondents who reported it, and those who did not are unlikely to be missing at random with respect to their involvement in the patient’s care. The survey also recorded the type of cancer and the respondent’s relationship to the diagnosed family member; neither was included in the models. For 396 of the 855 respondents with a family cancer history (46.3%), the diagnosed family member had died by the time of the survey, so for almost half of this subsample the satisfaction ratings are retrospective appraisals of care that ended in bereavement. Bereavement is a plausible determinant both of well-being and of how completed care is appraised. Adding bereavement to the cancer-affected model leaves the interaction and the direct contrast unchanged (Supplementary Table S7). Its own coefficient is close to zero, but the model already adjusts for current psychological distress and subjective health, through which any effect of bereavement plausibly operates; the analysis therefore indicates that the moderation is not an artefact of bereavement, rather than that bereavement is unrelated to well-being. Comorbidity was recorded on a multiple-response item, and the 70 respondents who declined to state their conditions were grouped with those reporting a condition, so that category is not a pure measure of diagnosed comorbidity. The cells defined by satisfaction with cancer care and recovery status are also small, with 28 recovered and 5 unrecovered cancer-affected households in the high-satisfaction category, so the moderation should be regarded as provisional. Finally, well-being was reported by respondents about themselves; it is used here as an indicator of well-being in a cancer-affected household, not as a measurement of the diagnosed family member’s well-being or of the household as a unit.
Conclusion
Unresolved subjective recovery and satisfaction with cancer care showed interacting associations with family well-being in Japan. Because the well-being advantage of high-quality cancer care was observed only in households whose recovery was achieved or in prospect, these cross-sectional findings are consistent with pairing reconstruction support with sustained access to family-centered cancer care rather than delivering them sequentially. Cancer-affected households merit explicit attention in local recovery planning, a proposition that warrants confirmation in prospective, longitudinal research.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/dmp.2026.10474.
Data availability statement
The datasets generated and/or analyzed during the current study are not publicly available due to privacy considerations but are available from the corresponding author, Dr. Akihiko Ozaki, upon reasonable request.
Acknowledgment
During the preparation of this manuscript, the authors used Claude (Anthropic), a large language model, to assist with language editing, proofreading, and improving the readability of the text. The authors reviewed and edited all AI-assisted output and take full responsibility for the content and conclusions of this work.
Author contribution
Conception and designing of the study: Ozaki and Murakami; Collection and assembly of data: Ozaki and Murakami; Data analysis and interpretation: All authors; Writing this paper: Hara and Ozaki; Critical revision of the paper: All authors. All the authors read the final draft and approved submission.
Funding statement
This study was funded by a grant from the Toyota Foundation (D17-R-0540). It was also funded by NPO Institute of Medical Care and Societal Health. The funding sponsors had no involvement in the study design, data collection, analysis, interpretation, report writing, or the decision to publish the results.
Competing interests
Dr. Ozaki received personal fees from MNES, Kyowa Kirin Inc., Becton, Dickinson and Company, Pfizer, Daiichi Sankyo Inc and Taiho Pharmaceutical Co., Ltd., outside the scope of the submitted work.
Ethics approval and consent to participate and consent for publication
The Minamisoma Municipal General Hospital Ethics Committee (approval 5-14) and Fukushima Medical University (approval 2019-198) granted ethical clearance for this study.
