This study extends the analyses conducted in the prior SHAPE study. As described in Finkelstein et al. (2021) SHAPE is a cross-sectional survey administered to 200 adult patients with advanced cancers (either advanced Stage IV solid cancer or an advanced stage of leukemia or lymphoma cancer) who were Singapore citizens or permanent residents, and with a likely prognosis of 1 year or less as determined by the primary treating oncologist. Other inclusion criteria included (a) age ≥ 21 years old, (b) diagnosed with advanced/stage IV solid cancer, leukemia or lymphoma, (c) mentally competent and aware of their condition. Patients were recruited between June 2018 and May 2019 at outpatient oncology clinics and inpatient wards at the Singapore General Hospital (SGH) and National Cancer Centre Singapore (NCCS). Initially, 293 patients were referred by oncologists, of whom 200 consented and completed the survey. Informed consent was obtained and the survey was administered by trained interviewers in either English or Mandarin, depending on the patient's preference, using tablets. In 2021, the study team reconsented 51 patients of the 69 patients still alive to have their medical and billing records up to March 2022 linked to survey responses (13 rejected, and 5 were uncontactable). Records of the 131 patients who died in 2021 or earlier were also linked. In total, 182 patients were included in the healthcare utilization/expenditure analysis and 195 patients were included in the survival analysis. By March 2022, when the study was completed, 142 of the 195 patients had died. The study was approved by the SingHealth Institutional Review Board (CIRB Ref. No: 2017/2181).

Healthcare utilization and expenditure were obtained from billing records of visits to specialist outpatient clinics, day surgery procedures, Accident & Emergency (A&E) visits, inpatient admissions, and pharmacy prescriptions. As public hospitals in Singapore are not expected to turn a profit, we used gross non-subsidized prices to proxy for the true cost to the health system. Our analysis focused on planned utilization received as more hopeful patients are expected to have greater planned utilization but may or may not have greater unplanned utilization. Planned utilization is defined as visits to outpatient oncology clinics, day surgeries and inpatient admissions not through the A&E department. A&E visits and admissions through the A&E are assumed to be unplanned. Information on patient's dates of death was extracted from medical records. Expenditures were reported in Singapore dollars (S$).

We calculated healthcare utilization and expenditures for the first 12 months after survey completion for those who survived at least 12 months after the survey or the last 12 months of life for those who died within 12 months after being surveyed (henceforth “12-month period around the survey”). In an additional analysis, we restricted the sample to the deceased only and calculated healthcare utilization and expenditures during their last 12 months of life. The former allows us to analyze the full sample but may not capture all aggressive treatments. The latter focuses on the period when more hopeful patients are most likely to pursue aggressive but marginally effective treatments.

Hope was assessed using the abbreviated version of the Herth Hope Index (HHI) (Herth, 1992). The Herth Hope Index (HHI) is an adapted from the Herth Hope Scale (HHS) and designed to be used in clinical settings. The instrument includes 12 items, grouped into three subgroups to measure inner sense of temporality and future, inner positive readiness and expectancy and interconnectedness with self and others (Herth, 1992). The items were graded on a 4-point scale labeled: Strongly disagree, Disagree, Agree, Strongly agree. Scores range from 12 to 48 with higher scores indicating higher degree of hopefulness. For the Mandarin speakers, we used a validated version obtained from a published psychometric evaluation of the translated instrument (Chan et al., 2012). As hope is a composite construct that may comprise many dimensions, we also examined two measures that are more closely associated with their medical condition/treatment (henceforth referred to as “illness-related hope”). These include expecting to live at least 2 years and believing that the primary intent of treatment is to be cancer free (Finkelstein et al., 2021). Patients' expectation of survival duration was assessed based on patients' answers to the question “Do you expect to be alive in …?” with options including “20 years”, “10 years”, “5 years”, “2 years”, “1 year” “9 months” and “6 months”. Treatment intent was based on the question “What do you think is the primary goal of your treatment regimen?” with response options of “Cure my illness so I will be cancer free at some point”, “Prolong my life”, “Manage my symptoms (e.g., control pain)”, “Don't know” and “Others”.

The survey also asked patients to report time since first diagnosis, primary cancer site (gastrointestinal, genitourinary, leukemia/lymphoma, lung, prostate and others), stage at first diagnosed (only for solid cancers), and self-reported health rating on the day of the interview. Socioeconomic data such as age, gender, education level and marital status were also included. The full instrument is included in Supplementary material Full Survey Instrument.

We separately estimated the effect of hope using regressions models that control for age, gender, educational attainment [primary school or below ( ≤ 6 years), secondary school (≤ 10 years), and post-secondary school (>10 years)], self-reported health rating of the day, cancer site, whether cancer was detected at an early stage, and years since cancer diagnosis. Robust standard errors were reported for all regressions.

Utilization, measured in terms of the number of visits or admissions, was estimated using a negative binomial regression. For expenditure, we used a generalized linear model (GLM) with log link and gamma-distributed errors. For each outcome of interest, we calculated the average marginal effects of hope as the average difference in predicted outcomes for a marginal change in HHI, a higher HHI quartile relative to the 1st HHI quartile (1st quartile: 15–35, 2nd quartile: 36–42, 3rd quartile: 43–46, 4th quartile: 47–48) to capture potential non-linearities. A similar approach was used to calculate the average marginal effects of illness-related hope (i.e. expecting to be alive in at least 2 more years and believing the primary intent of treatment is to become cancer-free).

To investigate for a potential survival advantage, we first examined Kaplan-Meier (KM) survival curves stratified by (1) HHI quartile subgroups and (2) presence of illness-related hope, and tested for differences in survival using the log-rank test. Adjusting for the same set of potential confounders as the healthcare utilization/expenditure analysis, we then fit a Cox proportional-hazards regression model, with separate hazard functions stratified by leukemia and non-leukemia cancers. This was because leukemia status failed the Schoenfeld residuals test of proportionality at the 5% significance level.

Our base case survival analysis conservatively assumes that self-rated health is a confounder influencing both hope and survival. However, one may argue for exclusion because self-rated health may also be an intermediate variable between hope and survival such that controlling for self-rated health absorbs the effects attributable to hope. Therefore, in additional analyses, we omitted self-rated health from the set of controls as a robustness check.

Table 1 describes the characteristics of our study sample. The final analytical sample included 195 participants for the survival analysis and 182 participants for the utilization/expenditure analysis. The mean age was 65.8 (SD: 65.7) and the majority were Chinese (80%), male (68%, SD: 46%), married (82%, SD: 38%) and diagnosed with solid cancer (90%). Close to two-thirds were at late stage of cancer (Stage IV) (62%) and close to half were initially diagnosed at an advanced stage (45%). The top three cancer sites were genitourinary (23%), lung (21%) and prostate (21%). By 2022, 142 (78%) participants had passed away, with close to half (46%) doing so within a year of the survey.

Within the 12-month period following the survey, patients incurred an average medical expenditure of S$101,816 (SD: S$109,663). On average, patients had 28.9 (SD: 15.6) planned visits/admissions costing S$83,940 (SD: S$94,808). In the last 12 months of life, decedents (n = 142) incurred an average medical expenditure of S$104,730 (SD: S$109,066). These patients averaged 29.1 (SD: 16.6) planned visits/admissions costing S$78,126 (SD: S$85,078).

The mean HHI score was 39.5 (SD: 7.5) out of 48. Cronbach's alpha for the study sample was 0.88 for both the English and Chinese versions. Nearly all patients (180, 92%) stated that they expected to live at least 2 more years and 80 (42%) participants believed that the primary goal of treatment was to become cancer-free. In reality, about 50% of participants passed away within 2 years of taking the survey. Figure 1 provides a visual representation of the overestimation of survival outcomes.

Table 2 reports the average marginal effects of hope on annual healthcare utilization and expenditure within the 12-months of taking the survey. Contrary to expectations, we did not find that HHI had a statistically significant association on the frequency of planned care or healthcare expenditure. The average marginal effect of HHI was small and not statistically significant. Quartile-specific estimates suggested that a non-linear relationship may be driving this null finding, but these estimates were generally not significantly different from the first quartile at the 5% level. Although difficult to explain, patients in the 3rd quartile of HHI (HHI 43–46) had the lowest total healthcare expenditure among the quartiles, S$29,157 lower on average compared to the 1st HHI quartile (95% confidence interval (CI) –S$60,255 to S$1,941, p-value = 0.07).

Patients who expected to live at least 2 more years had 6.6 more planned visits/admissions (23% increase; 95% CI 0.90 to 12.30, p-value = 0.02) on average compared to those who did not. Believing that the primary intent of treatment is to become cancer-free did not have a significant effect on planned visits/admissions. Neither measure of illness-related hope had a statistically significant association with either planned or total healthcare expenditure. However, point estimates suggested that patients who expected to live at least 2 more years had on average higher planned and total medical expenditure (S$11,112 and S$3,320 more respectively) than those who expected to live fewer than 2 years. Those who believe that the primary intent of treatment is to become cancer-free also had higher planned and total medical expenditure (S$3,069 and S$11,147 more respectively) than those who did not have this belief.

In efforts to provide further insight into the results, we restricted the analysis to the decedents-only sample and in the last 12 months of life. The results are generally similar with those based on the full sample (see Supplementary Table A1). HHI did not have a statistically significant association with frequency of healthcare utilization and expenditure, although point estimates for HHI quartiles showed a more consistent trend of fewer planned visits/admissions and expenditure among more hopeful patients, which is contrary to our initial hypothesis. Neither measure of illness-related hope was statistically associated with planned healthcare utilization or expenditure. Nevertheless, point estimates remained positive implying that more optimistic patients were more likely to utilize planned care, which is consistent with expectations. Patients who believed that the primary intent of treatment was to become cancer-free spent significantly more in total medical expenditure (S$30,712, 95% CI S$3,143 to S$58,282, p-value = 0.03) compared to those who did not believe.

Figure 2 presents the KM survival curve of patients, stratified by measures of hope and optimism. Consistent with our initial hypothesis, survival is trending upward for the first three quartiles. However, contrary to expectations, it then drops for the fourth quartile. Regardless, survival functions were not statistically different across HHI quartiles (p-value = 0.38), as determined by the log-rank test. The survival of those who expected to live at least 2 more years was higher than those who did not (p-value = 0.02), but not those who believed that the primary intent of treatment is to become cancer-free (p-value = 0.59).

Adjusting for confounders in the stratified Cox regression did not change our results (Table 2). Only higher survival expectations had a significant association with actual survival. Those who expected to live at least 2 more years had a 40% lower hazard of death than those who did not (hazard ratio (HR): 0.592, 95% CI 0.355 to 0.987, p-value = 0.04). Excluding self-rated health as a covariate in the Cox regression model showed that one point increase in HHI was associated with 2.2% points lower hazard of death (HR: 0.978, 95% CI 0.956 to 0.999, p-value = 0.03) and that the 3rd HHI quartile had 36.0% lower hazard of death as compared to the 1st HHI quartile HR: 0.640, 95% CI 0.396 to 1.035, p-value = 0.08) (Supplementary Table A2). There was no change to the estimated effects of biases (Supplementary Table A2).

This study investigated the relationship between hope, health seeking behavior and survival among advanced cancer patients. Contrary to expectations, we did not find a clear positive relationship between a broad measure of hope and these outcomes. We found that the 4th HHI quartile (those with HHI scores of the maximum possible or one point below) incurred lower planned and total healthcare expenditure and also had lower survival odds compared to the other quartiles, although the differences were not statistically significant and may simply be due to chance. These results are consistent with prior studies. Petticrew et al. examined 26 studies focusing on two aspects of hope, including fighting spirit and helplessness/hopelessness and found no significant associations with survival (Petticrew et al., 2002). We did find that excluding self-rated health as a covariate showed that higher hope was significantly associated with greater survival, but our study design did not allow us to discern if self-rated health is an intermediate variable that should be excluded, or a confounder that should be included.

The lack of associations may result because hope, as measured by the HHI, may be more distal compared to our measures of illness-related hope. We found some evidence that illness-related hope was associated with greater healthcare utilization/spending and a greater survival advantage. Patients who expected to live for at least 2 more years had more frequent planned healthcare utilization in the 12-month period around the survey, although this did not translate into significantly higher healthcare expenditure. Furthermore, patients who believed that the primary intent of treatment is to become cancer-free spent S$30,712 more on total healthcare in their last 12 months of life than those who did not.

Although our study did not capture communication between providers and patients, other studies reveal that providers are often hesitant to convey negative information to patients so as not to dampen hope (Weeks et al., 1998). Our results suggest these concerns may be well founded. We found that patients who expected to live for at least 2 more years had greater survival than those who did not. This finding is consistent with Lee et al. who also found that optimistic expectations were associated with better survival (Lee et al., 2003). Yet, this survival increase was not without cost as we also showed that an unintended consequence of this potentially unfounded optimism was to increase healthcare utilization and expenditures at end-of-life.

This analysis has several limitations that need to be considered in the context of these results. First, we focused on measures of hope at a single point in time. Yet Solano et al. found that hope changes with illness (Solano et al., 2016). A review by Kylma et al. further showed that hope is related to temporal factors such as pain and quality of life (Kylm et al., 2009). Prognostic beliefs have also been shown to change over time (Ozdemir et al., 2023) and may be influenced by both health status and prior treatment decisions (Finkelstein et al., 2021). For example, Echarte et al. proposed that self-deception entails a lower cognitive load than accepting the truth and argues that, as a result, patients can become more hopeful as illness progresses (Echarte et al., 2016). Relatedly, our hypothesis assumes that patients' primary hope is for a cure, or at least for significant life extension. This is likely true upon diagnosis and consistent with our prior findings showing the relationship between hope and several biases. Nevertheless as illness progresses, hope may shift toward more obtainable goals, such as finalizing affairs and being cared for at a place of choice (Duggleby and Wright, 2004; Clayton et al., 2005). Similarly, availability, accessibility and quality of end-of-life care, including conversations about palliative care, may shift patients' hope and intention away from aggressive care and life extension (Earle et al., 2008).

Second, we were also unable to rule out reverse causality between hope and healthcare utilization/spending. This could result, for example, if frequent hospitalizations induce the illusion of treatment success which then influences decisions for further treatment. Conversely, hope may decrease after rounds of unsuccessful treatments, which could lead to decisions to suspend efforts to extend life.

Third, healthcare utilization/expenditure and survival duration may be censored for patients who stopped receiving treatment at SGH or NCCS because they transferred to a private care setting. Although we were told by the clinicians that these cases are rare, our estimates could potentially be biased if more hopeful patients are systematically more likely to transfer out of SGH or NCCS. However, we suspect this is unlikely to be an issue as NCCS cares for 70–80% of cancer patients in Singapore and we did not find any significant association between HHI and the likelihood of a dying outside SingHealth network of care providers.

Finally, our measures of hope are imperfect. HHI, which is one of several available hope indices (Bryant and Harrison, 2015), is a broad measure of hope encompassing three dimensions: temporality and future, positive readiness and expectancy, and interconnectedness (Herth, 1992). It is possible that results would differ if an alternative measure of hope were employed. On the other hand, hope indicators more closely related to participants' condition could be the result of biases or misinformation, potentially arising from poor communication with providers (Weeks et al., 1998). We were unable to capture what information was conveyed by the providers to the patients. Due to litigation risk resulting from nondisclosure (Austin, 2019), it is likely that patients were informed about prognosis and treatment intent. However, the extent of disclosure and whether or not patients truly understood what was conveyed is unclear.