The study is based on a prospective, population-based cohort established in Denmark in 2012 and followed to 2022. The Danish health care system is tax-funded. Services are free of charge for residents, and GPs serve as gate-keepers to secondary care, including palliative care and psychiatric services (22). However, 40% of the costs of psychotherapy by a psychologist is self-paid after referral from a GP (23). Seven GP talk therapy sessions are free of charge per year, and GPs attending peer supervision may provide 20-min sessions focusing on mental health issues (24).

In 2012, we obtained register-based information on all patients receiving drug reimbursement due to terminal illness, i.e. patients with a life-expectancy of only a few weeks or months (25). Weekly we sent letters to patients newly registered with drug reimbursement and asked for the patient's closest relative to complete the questionnaire (26). Enrolled relatives completed a questionnaire at the time of inclusion (T0), at 6 months after bereavement (T1), and at 3 years after bereavement (T2), and the outcome variables were followed in individual time periods for up to seven years after T2 (Figure 1). The study period ended December 31, 2022. Information regarding the relation to the patient (partner, adult child, other) was collected in the baseline questionnaire at T0.

The Prolonged Grief-13 (PG-13) scale (22) was used to measure grief symptoms at T0, T1, and T2. At baseline (T0), we used an adapted version of the scale (26) including “loss” changed to “illness” to correspond to the time prior to the patient's death in line with prior studies (27).

We have previously identified five specific grief trajectories based on the PG-13 measures at T0, T1, T2 using a semi-parametric group-based trajectory model (GBTM) for repeated measurements (28, 29). Details of the method have been described elsewhere (23, 24).

A unique personal registration number is granted to all Danish citizens. This number allows for linkage between individual-level records in Danish registers and the collected questionnaire data, which was processed at Statistics Denmark.

Information on the age of the relative, gender, educational level according to the UNESCO international standard of classification [low ( ≤ 10 years), intermediate (>10 and ≤ 15 years), high (>15 years)] (33). The Danish National Patient Registry (34) provided information of diagnoses registered in connection with a contact to a hospital, which allowed us to adjust for the presence of at least one of the somatic diseases in the Charlson Comorbidity Index (CCI) (35).

For all the analyses, we described the association between the five grief trajectories and outcome measures with those in the LGT as the reference group. Relatives contributed with at risk time up until date of death, date of emigration or December 31, 2022, whichever came first.

To examine the incidence rate ratios (IRRs) for GP contacts (daytime face-to-face and out-of-hours) according to grief trajectories (LGT as refence) we applied negative binomial regression analyses to account for overdispersion. Repeated yearly measurements were addressed using cluster robust variance estimation (36) and follow-up time was used as the offset. Cluster robust variance estimation was used to account for the repeated yearly measurements within each relative.

To examine the odds ratios (ORs) for the dichotomous outcomes mental health contacts and medicine use according to grief trajectories (LGT as reference) we applied logistic regression models. The dichotomous outcomes of mental health contacts and medicine redemptions were analyzed using logistic regression models with follow-up time used as the offset.

Finally, to examine the hazard ratios (HRs) for death according to grief trajectories (LGT as refence) we applied Cox proportional hazards models with age chosen as the underlying time scale. The proportional hazards assumption was assessed graphically using log-minus-log plots and no apparent violation was observed.

The negative binomial regression models yielded incidence rate ratios (IRRs) of yearly GP contacts, the logistic regression models yielded odds ratios (ORs) of medication use and mental health services within the seven-year follow-up period, and the Cox proportional regression model yielded hazard ratios (HRs) of mortality. All regression models were adjusted for the a-priori chosen covariates of age, gender, personal relation to deceased, education and CCI. All estimates were presented with 95% confidence intervals (CIs) and ratio estimates were considered statistically significant if 1 was not included in the CI. All analyses were done using Stata 18 (StataCorp, Texas, USA).

The study population of 1,735 persons included predominantly females (71%). In all, 1,138 were partners (66%), 476 (27%) were adult children and 121 (7%) had another relation to the patient. The mean age was 62 years, 449 (26%) had a low education and 17% had one or more comorbid diseases (CCI≥1) (Table 1).

Relatives in other GT groups than the reference group (LGT) had more GP contacts during the first years of follow-up and the difference seemed to level out toward 10 years of follow-up (Figures 2, 3). Relatives in the LaGT had more overall contacts 4 and 5 years after bereavement, but no statistically significant excess face-to-face contacts compared to relatives in the LGT.

Relatives in the HDGT had significantly more total contacts in year five to seven and more face-to-face contacts in year six after bereavement, whereas those in the HGT had significantly more GP daytime contacts (face-to-face contact and total contacts) in year four to seven after bereavement compared to the LGT (Figures 2, 3). From year eight to ten after bereavement, no significant difference in contact pattern was found between any of the grief trajectories.

The pattern of initial contact with out-of-hours services showed that in year four after bereavement, the MDGT had significantly more contacts than the LGT, in year five the HDGT had more contacts and in year five and six, the HGT had significantly more contacts than the LGT (Figure 4).

In the adjusted logistic regression analysis, the HGT was associated with more mental health services [OR = 2.86 (95% CI: 1.58;5.19)], use of antidepressants [OR = 5.63 (95%CI: 3.52;9.01)] and anxiolytics and sedatives [OR = 2.60 (95%CI: 1.63;4.14)] compared to the LGT (Table 2).

Use of antidepressants was also higher in the LaGT [OR = 2.46 (95% CI: 1.53;3.97)] and HDGT [OR = 1.87 (95%CI: 1.29;2.70)] and use of anxiolytics and sedatives was higher in the HDGT [OR = 1.77 (95%CI: 1.26;2.48)] compared to the LGT (Table 2).

In total, 186 (10.7%) died during 3–10 years after the patient's death, and the HGT was associated with mortality [HR = 1.88 (95% CI: 1.11; 3.21)] compared to the LGT (Table 3).

Bereavement has consistently been associated with increased use of health care and psychotropic medication (4, 5, 18). The present study establishes that individuals most likely to receive excess health care services beyond 3 years after bereavement were those with high levels of persistent grief symptoms. The long-term findings regarding health care are likely to be impacted by events such as illness of the relative or other losses during the ten-year period after the patient's death. However, such events may also impact the other grief trajectory groups.

A prior study of health care use before bereavement showed that the HGT was associated with more contacts to GP, medication use, and use of mental health services (24). The relatives in the HGT may have predisposing vulnerabilities and challenges and that severe illness and bereavement pose a high level of distress that they find difficult to handle. Furthermore, relatives with HGT had a lower educational level for both spouses and adult children of the ill person (23). Hence, prior mental health conditions and low educational level are risk factors for long-term psychological distress in bereaved relatives. Nevertheless, the findings of the present study may be connected with confounding by indication as the persons in the HGT had higher health care use prior to bereavement, but the findings underline that relatives in the HGT are persons who are likely to need additional support from the health care system.

The contact pattern to out-of-hours GP services was diverging. These services are intended for acute situations and available free of charge to all citizens in Denmark (37). Hence, it is at first sight encouraging that persistent grief symptoms may not significantly affect the long-term use of these services as this could indicate that the supportive services provided during daytime sufficiently support the relatives. However, the connection between daytime consultation and out-of-hours use may be complex and those with need for mental health support might call helplines for mental health. A recent study showed that the use of out-of-hours GP services among citizens above 75 years of age was not lower if the patient had more daytime consultations in general practice (38), and personal characteristics rather than organizational factors were suggested to affect the use of out-of-hours GP services. Hence, out-of-hours contact may not be relevant health care for the participants.

Regarding psychotropic medication, the use of antidepressant and anxiolytics and sedatives was significantly higher in the HDGT and HGT than in the LGT. Approximately 18% used psychotropic medication in the first year after bereavement (39). Our results indicate that early grief patterns may predict health care use is a new finding that extends prior knowledge of higher medication use in bereaved compared to non-bereaved (4, 18). Hence, a specific group of relatives may be recognizable for intervention, especially as the HGT has been found to be associated with higher medication use already before bereavement (24).

We found that relatives with HGT had a higher use of any mental health service after bereavement. Hence, the available services seem to be directed at those with the most severe symptoms. However, the persistent need for mental health service over a long-term period could indicate that the existing interventions may not be sufficient to address the needs of bereaved relatives. Also, the vulnerability of the relatives in the HGT group may also be related to prior health and mental health conditions contributing to their need for individually adapted mental health services.

Talk therapy sessions with the GP and private-practice psychologists early in the bereavement period have been found to reduce the risk of suicide, deliberate self-harm, and reduce psychiatric hospitalization in bereaved persons (40). The HGT was associated with GP talk therapy but not psychotherapy by a psychologist among relatives before the death (24). Thus, general practice holds an important task to support those in the HGT, who are more likely to have a low educational level and may not have the economic or mental resources to attend therapy with a psychologist. In recent years, more attention has been drawn at relatives and new knowledge has emerged to inform health professionals, including a supportive intervention in general practice using a dialogue questionnaire (41) and the integrated process model of loss and grief to support the understanding of grief in general (1).

We found that relatives with HGT had a higher mortality compared to relatives with LGT. Previous studies did not compare mortality according to symptoms levels, but excess mortality has been found for bereaved compared to non-bereaved persons in most (3, 17), but not all previous studies (18). Bereaved relatives constitute a heterogeneous group with different risk factors for experiencing adverse bereavement outcomes (3). Caregiving for a close relative have been suggested to buffer the stress reaction in relatives (19). However, if the relative suffers from distress due to their own physical illness this may add to their vulnerability and grief symptoms. Development of persistent high grief symptoms may be due to several risk factors, and our findings support that comorbidity is likely to play an important role as an intrapersonal risk factor for bereavement outcome (12).

The main advantages of the current study are the follow-up time of 10 years, the longitudinal study design of grief symptom assessment at three time points in 1,735 relatives, and the combination with valid register data on health care services. The Danish registers are considered almost complete, and the data can be linked precisely (42). We used grief trajectories based on repeated grief measurements as exposure, which can be considered a strength as persistent grief reactions were included. The association between having persistent high levels of grief symptoms and higher use of health care services may be considered to be confounding by indication as those in the exposure group HGT with the highest symptom level may have an a priori higher use of health care due to symptoms.

The study was limited by the number of participants and thus lacked power to investigate for instance the different types of mental health services. Also, the follow-up of 10 years may be seen as a limitation to reveal long-term impacts of bereavement on mortality. The study included no comparison group from the background population, which was a limitation. Beyond seven years of bereavement, the association of higher use of primary health care in the HGT seemed to level out. This may indicate that the impact of bereavement diminishes over time. However, it could also be due to the group size as a small group of relatives in the HGT could potentially have had many contacts and this effect would diminish if they had died during follow-up.

The study population was younger and had a higher educational level than non-participants (18, 23). As those with high education may have lower symptom levels, the prevalence of high grief symptoms might have been underestimated in the present study. Still, we believe that the results may be generalizable to bereaved relatives in Denmark and in countries with similar health care systems.