In this study, we demonstrate substantial differences in mortality among Medicare patients hospitalized with injuries depending on the follow-up period used. In-hospital mortality, a very commonly used outcome measure in injury research, was less than half the 30-day mortality (following date of admission), suggesting that a substantial portion of hospitalized injured elders will die shortly after being discharged from the hospital. These findings also demonstrate significant limitations with geriatric injury studies using in-hospital mortality as the primary outcome measure. Our results also demonstrate that the change in mortality rates over time appears to stabilize by six months after injury, suggesting that this time period represents the preferred duration of follow-up, although 60 days may be a reasonable proxy, particularly for seriously injured patients.
Previous studies have shown the inadequacy of relying upon in-hospital mortality as a primary outcome measure. Jencks et al demonstrated that conclusions about outcomes for elders admitted in different regions of the country for four medical diagnoses differed based on whether in-hospital or 30-day mortality was measured.13
A study of injured elders by Gorra et al demonstrated similar variation between regions based on when the mortality outcome was measured.14
Mullins et al found that in an all-ages cohort of admitted injured patients, substantially more patients died within 30 days of being discharged from the hospital alive than died in-hospital.5
This effect was most prominent among elderly patients, particularly those identified as having a cause of death related to trauma. Our study reproduces this finding of many deaths closely following discharge in an elderly population.
This balance between in-hospital and early post-discharge mortality may be affected by factors at the patient, provider, and community level. Patient level factors include disease processes which may be incompletely treated and more or less appropriate for palliation outside the acute care hospital, family resources for home care, and end-of-life preferences. Provider-level factors may include trauma center designation, availability of case managers, and provider thoughts about futile care. Community-level factors may include established regional treatment practices and the availability of skilled nursing and rehabilitation facilities. Recognizing the multi-dimensional causes of variation in hospital stays, we believe our results and those of others support evaluating outcomes at fixed intervals after injury when possible.
Given the agreement of our findings with previous authors that in-hospital mortality is a suboptimal outcome measure for injured elders, we sought to find the shortest appropriate fixed follow-up period after injury. One approach would have been to follow patients until the mortality rate for the injured cohort returned to that seen in the general population. This approach is unworkable, however, as previous studies by Gubler et al and McGwin et al both showed that mortality rates for injured Medicare patients never returned to baseline, even through five to six years of follow-up.1,2
Another approach would have been to follow patients through the episode of care related to their injury, linking claims that appeared to be temporally and anatomically related, as done by Lestina et al for a cohort of injured managed care patients.15
The limitation of this approach to the issue of mortality assessment is that episodes of care will be quite different for patients with fatal versus non-fatal injuries. Our approach was to consider the period after injury during which the daily mortality rate (hazard function) was elevated as the high-risk period through which surveillance should continue.
We found that the daily mortality rate did not change from 180 to 365 days post-injury. These results are consistent with Gubler’s et al and McGwin et al’s findings. While neither author reported exactly how long their initial period of declining mortality rates persisted, examination of survival curves from both studies suggests that it was less than six months after injury. Our results were also consistent with those of Wunsch et al, who found that while cumulative mortality in hospitalized Medicare patients was elevated at three years of follow-up when compared with matched non-hospitalized controls, this effect was most pronounced in the first six months after hospitalization.16
This effect was greatest in those who required mechanical ventilation in an intensive care unit.
Although our daily mortality rates had stabilized by six months post-injury, 60 days appeared to be an acceptable alternative follow-up interval, with 89% of the fall in daily mortality rate accounted for by 60 days. Even greater capture of post-injury changes in mortality rates would be achieved by extending follow-up to 90 days, which would capture 95% of the excess daily mortality rate. Both in-hospital and 30-day mortality provide incomplete follow-up periods, as the daily mortality rate continued to fall within the 30–60 day window.
To assess applicability to a more seriously injured population, such as that seen in trauma registries, we analyzed the subgroup with an ICISS ≤ 0.9. The seriously injured experienced the highest early mortality as well as the earliest decline to a steady daily rate of mortality after injury. While this decline was not so precipitous that we would advocate using 30-day follow-up as an ideal measure in a seriously injured population, 60-day surveillance should be reliable. We used prior nursing home residence as another marker of poorer baseline functional and medical status; this group also displayed similar results of an overall elevated cumulative mortality but a similar pattern of decline in the mortality rate. This finding suggests that these conclusions can be applied equally to studies involving community dwelling or institutionalized elders.
Hip fractures comprised more than a third of our cohort. Our cumulative mortalities () are similar to those reported by Brauer et al in a recent observational study of a nationwide Medicare sample.17
Our 30-day mortality of 10.3% is within the range she reported for 2002 of 6% for women and 12% for men. Our one-year mortality of 30.6% was also similar to her estimates of 23% for women and 35% for men. This supports our conclusion that in-hospital mortality is a poor marker of overall outcomes for the hip-fracture population as well.
Previous studies have demonstrated the influence of medical co-morbidities on trauma outcomes. In a case-control study of adult trauma patients of all ages, Morris et al showed significantly more comorbidities in those who died versus surviving controls matched by age and injury severity.18
Hollis et al demonstrated increased odds of death in injured patients with medical co-morbidities when stratified by age, although this effect was only sustained in mildly and moderately injured subsets.19
We compared the group with a high co-morbidity burden (Charlson score ≥3) to the entire cohort. This group had a substantially higher mortality at most time points, supporting the conclusion of Hollis et al that pre-existing medical conditions influence trauma outcomes. While overall mortality was higher among those with comorbidities, the pattern and timing of their deaths after injury were similar to the overall cohort, again supporting a stable daily mortality rate within a follow-up interval between 60 days and six months. This subgroup exhibited survival curves and daily mortality rates similar to the prior skilled nursing facility residents.
Our findings have important implications for both the researcher and the geriatrician. For the researcher, they show that studies using in-hospital mortality will be heavily biased by factors influencing whether death occurs in-hospital or shortly after discharge. Investigators should make every effort to identify and control for these factors, define mortality outcomes at a 60-day to six-month time-point after injury, or use outcomes based on survival time rather than dichotomous measures of mortality. For clinicians, our findings highlight the need for greater planning for post-discharge care to prevent post-discharge death when possible. Knowing these survival patterns will help geriatricians counsel their patients on treatment plans that agree with their end of life preferences.
Our cohort consisted of elders hospitalized with injury diagnoses. Therefore, injured elders who did not survive to hospital admission or who were discharged home from an outpatient setting (i.e. the emergency department or clinic) were not included in the sample. Inclusion of these patients may have changed our findings, although it is uncommon for an elder at high-risk to be released without at least a period of observation in the hospital. Another piece of information missing from our data was mechanism of injury. While the dataset did contain ICD-9 mechanism of injury codes (e-codes), these were reported in only 29% of cases, and were therefore considered to be unreliable. Furthermore, we are forced to assume that the injury occurred on or just before the date of admission.
While mortality is an easily defined outcome, it does not tell the full story of the sequelae of geriatric injury. Even injuries which are not fatal may have significant effects on functional outcomes (e.g., return to independent living, ability to ambulate, self-sufficiency with activities of daily living). We did not examine non-mortality or functional outcomes, as reliable markers of function were not available in the database. Discharge disposition (e.g., to an inpatient rehabilitation facility or skilled nursing facility) was inconsistently reported. Similarly, the presence or absence of nursing facility bills may be unreliable, as Medicare only pays for inpatient rehabilitation and skilled nursing facilities, but not for other types of institutions for the elderly. These are inherent limitations of outcomes research based on administrative billing databases. A strength of Medicare data is its accuracy in reporting the date of death, as it is based on daily reports from the Social Security Administration.
Using a regional dataset may also limit generalizability. For example, elders in the Pacific Northwest are racially homogenous and may not be comparable to elder populations in other regions of the US. Another limitation is the lack of data from the 19% of the Medicare entitled population covered by Medicare managed care organizations, a percentage that is even higher in the West.8
These patients may have different baseline characteristics than the fee-for-service Medicare population which constituted our sample.