The deceased accounted for 13% of annual health spending for people 65 and older, but only 2.8% was spent on those who were more than 50% likely to die, according to our machine learning model. While the average health expenditure of a survivor per day was ten times higher than that of a survivor, when compared to an equally vulnerable population of survivors, the average expenditure of a survivor was only 2.5 times higher . The main strength of the study is the availability of data for the entire population, including rich health care and socio-demographic predictor data and 97% registry coverage of all health care expenditure.12, as well as the inclusion of community care in addition to treatment. Since health care expenditure in Denmark is tax-financed, differential insurance coverage and rates will not be artifacts of difference. However, individual-level expenditure data may be somewhat misrepresented: hospital costs are DRG rates that are averages and may not fully correspond to the actual cost of treatment, and nursing home and home care costs for individual The calculation of level expenses involves some amount of estimation and attribution. This study is concerned only with expected mortality at baseline, which may arguably be a limited indicator of the cost-efficiency of health care spending, and may take into account other measurements such as quality-adjusted life years.
The distribution of estimated mortality is similar to the estimate2 For American Medicare enrollees. Including a wider range of individual characteristics did not materially improve prediction, as our AUC is essentially the same as that of the Medicare study – a result that compares reasonably well to other studies. .6,7,13,14,15,16, especially considering the relatively broad time horizon of the prediction for our study. The very low proportion with the high estimated mortality may be due to essential randomization in mortality, accumulation of health-affecting events after the start of follow-up, or deficiencies in available data. But while we can absolutely point to health indicators that weren’t available to study, there are indications10,17,18 These may not improve the prediction of mortality that much.
The bulk of treatment costs are concentrated on under-estimated mortality in a pattern similar to Eenav et al.2, In contrast, care-related costs are concentrated at high mortality rates and increase more markedly with increasing mortality rates, whereas treatment costs among the dead actually decrease to an estimated mortality rate of about 30%. This is not surprising – the estimated mortality rate is a proxy for frailty and thus for the need for communal care, and the need for care is less likely to change as a result of health-affecting events during follow-up. . It is interesting that we see a decline in treatment-related expenditure per day alive for the deceased with estimated mortality. This was not observed for the US population and may reflect the different medical culture in Denmark and the US, but different prediction algorithms may also be part of the explanation – treatment expenditure decreases with age in the Danish deceased.1 1And if a higher estimated mortality rate is more reflective of age and vulnerabilities in our algorithm than in the US data, this may explain the difference.
At similar estimated mortality rates, there is little difference between the expenditure related to living care per day of deceased and survivors. The medical costs of the dead are much higher than those of the survivors, although the difference is small at high mortality rates. This pattern can be explained by the passage of time – by the time a person dies, their health is likely to have deteriorated since their condition on entry, and it seems likely that a person dying at a lower predicted mortality rate has done something The dramatic health experienced would have been an event requiring treatment, whereas death at a high predicted mortality rate may have been a more direct continuation of a pattern already established at the time of admission. Furthermore, a person with a lower predicted mortality rate may be a better candidate for treatment because of being less frail. But to the extent that the difference between survivors and dead at the same mortality rate is not due to curveball events, it can be viewed as the “true” cost of dying.
Thus, almost all health expenditure occurs in situations where there is a reasonable expectation that the patient may survive, and so the concept of “cost of dying” is confused with vulnerabilities: we spend more on vulnerabilities, And the vulnerable are more likely to die – but are not certain to do so, at least within the relevant time frame. This inherent flaw, in the last year of life in Denmark, is driven by a high estimated one-year mortality rate of 39% of health care expenditure, an estimate consistent with that of US Medicare enrollees.2, The idea of the potential to cut health care expenditures at the end of life is enticing, and it seems possible to find groups that might benefit from a shift to a palliative course of treatment. Nevertheless, our results, along with those of our model paper, add to a list of arguments why reducing health care spending by cutting the cost of dying may be misleading. Proportion of end-of-life expenses reported lower than previously reported1Deceased make up a relatively small portion of high-cost individuals3Rising levels of demand increase health care costs in an aging population at least as much as the cost of dying19And the high end-of-life costs seem to be driven more by multimorbidity than by last-ditch lifesaving efforts.1,11,20, Our study design does not touch upon the question of individual treatment effects – whether specific treatments improve survival for specific individuals – and it may be that better methods than ours could detect high-mortality subgroups, but It seems unlikely for such subgroups to be large enough where cost reductions may matter at the scale of the national budget.