Rate and risk factors of in-hospital and early post-discharge mortality in patients admitted to an internal medicine ward

Discussion

This is the first prospective study conducted in an internal medicine setting to look at determinants and comparison of both in-hospital and early post-discharge mortality by including biological and non-biological variables that may affect these outcomes. Although previous studies have analysed this same issue (Table 1), only three of them were prospective, not based on administrative data, and extended the observation to the early post-discharge period.^5,6,9 Moreover, a comparison of factors affecting in-hospital vs early post-discharge mortality has never been reported before. For these reasons, our study fills some relevant knowledge gaps concerning early mortality in the internal medicine setting which deals with clinically complex patients.³² We found some important factors associated with early mortality, such as older age, high dependency, and malnutrition that could be targeted for personalising ad hoc interventions, possibly improving patient outcomes and resource rationalisation.

Before performing any comparison of factors affecting mortality across the available studies, it is worth highlighting the wide heterogeneity of the different study populations. Our 1,451 patients had a median age of 80 years, and the in-hospital mortality rate (6.5%) was a little lower than the median mortality (7.9%) in the studies reported in Table 1. Notably, we excluded those patients with an expected prognosis of <24 hours, and this may have affected our overall in-hospital mortality rate. However, it should be noted that previous studies^1–16 are only partly comparable to ours, due to differences in methodology, patient age and enrolment criteria. For example, in the study by Sanclemente et al, in which no restriction based on prognosis was stipulated, roughly one-third of the deaths occurred within 48 hours after admission.¹ One study specifically focused on weekdays vs weekend hospital admissions.⁴ Other studies focused on specific age ranges, including age ≥65 years⁶ or ≥80 years.⁹ Although our only age limit was adulthood (≥18 years), most of our patients were >60 years old. Regarding frailty, another key clinical variable, our results were similar to those reported by Joosten,³⁸ who, however, studied the prevalence of frailty in a geriatric population and using a different scale.

Ageing is a well-established risk factor in both in-hospital and 4-month mortality,^{3,4,6,13,15,16} and, as expected, we confirmed these figures in our study. The detrimental effect of ageing is still significant at the 4-month follow up, possibly reflecting the poor homeostatic resources of the elderly.

Data regarding multimorbidity as a determinant of early mortality are heterogeneous and partly conflicting. Indeed, multimorbidity is common in internal medicine, and is generally associated with worse outcomes, as shown in the previous studies.^1–16 Our population reported a CIRS comorbidity or severity index >3 in most cases, similar to the study by Mannucci, who also used CIRS to assess multimorbidity.⁶ However, other studies reported multimorbidity using different tools. Multimorbidity was sometimes defined as the association of two or more diseases,^8,9,16 or else instances of specific disease clustering were taken into account,³ with a particular focus on neoplastic diseases or major clinical syndromes, including anaemia, chronic obstructive pulmonary disease, chronic heart failure, liver cirrhosis and chronic kidney disease.^{1,3,6,8,9,13,15,16} In our study, multimorbidity, as assessed by CIRS, was not correlated with either in-hospital or post-discharge mortality. This is probably because, as already shown in the SMAC study,³⁹ upon admission, the majority of patients had a CIRS comorbidity index >3, and hence this factor was far too common to highlight heterogeneity within the population. In other studies, multimorbidity had different impacts on early mortality, sometimes being significantly associated, as in Smolin et al, Han et al and Aljishi et al (according to the Charlson Comorbidity Index),^7,8,15 or had no significant effect, as in Perysinaki et al (according to the simple enumeration of diseases).⁹ Altogether, these data highlight the need for harmonisation in the reporting of multimorbidity and for further prospective studies to assess the effect of multimorbidity on mortality.

Polypharmacy, defined as the use of >5 medications, was not found to be associated with mortality. A landmark study focusing on this issue in internal medicine instead found an opposite result,⁶ attributing this finding to possible drug interactions. This apparent discrepancy may have different explanations. According to a recent meta-analysis, only excessive polypharmacy, defined as the use of 10 or more medications, was associated with death.⁴⁰ As the number of prescribed drugs increases with increasing number of chronic illnesses, in our study we preferred to use the internationally recognised definition of polypharmacy as a categorical variable, rather than a continuous one. Finally, as our hospital is an academic, tertiary referral centre, medication deprescription is part of the routine clinical practice,⁴¹ and this may have lowered potentially threatening drug interactions. At our hospital, both at the time of admission and discharge, a medication reconciliation is performed, ie, a careful assessment of all the medications taken by the patient, also discussed with their relatives, spouse, or caregiver. Indeed, potential drug–drug interactions and the risk–benefit profile is always considered in this process.

Many of the previously published studies frequently reported physiological and laboratory parameters, but gave relatively poor consideration to dependency as a significant variable that could influence patient outcomes. In our study, the Barthel index was lower than 60 in roughly 20% of all patients, meaning that one in five patients had at least moderate dependency. This finding appears to be even more significant if we consider that a dependency status turned out to be a significant determinant of both in-hospital (OR 2.1) and 4-month post-discharge mortality (OR 1.6). While one might expect age to be a strong determinant of mortality, the impact of dependency on mortality is not that obvious. One possible reason for this is that highly dependent patients at admission to hospital could undergo a progressive change in their daily habits that could also affect the clinical outcome. Similarly, Mannucci et al,⁶ who reported a Barthel index lower than 50 in about 16% of all patients enrolled in their registry, highlighted the importance of a routine assessment of the functional status of elderly patients using appropriate tools such as the Barthel index. The authors suggested that the traditional approach based on history and physical examination might be inadequate for identifying subtle impairments in daily life activities that we found had such a significant impact on the early mortality rate. Notably, in the prospective study by Perysinaki et al,⁹ although this was restricted to octogenarians and nonagenarians in the multivariate analysis, functional dependence was the strongest significantly independent predictor of mortality (HR 7.40), up to the 6-month follow-up. In addition, functionally independent individuals and those living at home with no need for caregivers lived longer. This points to the need to prevent disability⁴² and ensure the correct allocation of resources for frailer patients, especially outside of the hospital.

A low BMI was another key factor associated with 4-month mortality, but not with in-hospital mortality. BMI is frequently used as screening parameter of the nutritional status of hospitalised patients, and a low BMI underlies malnutrition. In this regard, our results are in line with those of another study⁴³ that already reported the association between higher early mortality rates and malnutrition, as measured with using either BMI or nutritional assessment scores. However, what is novel here is the lack of association between malnutrition and in-hospital mortality, while malnutrition affected the likelihood of death after discharge. In our hospital, all malnourished patients are assessed and treated by a physician nutrition specialist (ie a board-certified physician with clinical experience in nutrition therapy), and this might have influenced the in-hospital mortality outcome, as already demonstrated in a recent, large study assessing nutritional support in hospitalised patients.⁴⁴ Hence, our results seem to point to malnutrition as a main determinant of death when the patient is re-allocated at home early after a major health event (ie recent hospitalisation). Early after discharge, patients may be particularly frail and could have poor nutritional support, hence malnutrition should be prevented starting from the first day of hospitalisation and should be continued thereafter in order to avoid future detrimental effects, such as hospital readmission and death.⁴⁵

The last factor found to be associated with 4-month mortality was the length of stay. Indeed, a longer length of stay reflects the severity of the illness and more significant resource requirements,⁴⁶ and this may affect the subsequent risk of death. Notably, a previous study showed an opposite result, with a positive association between the in-hospital standardised mortality ratio and the length of hospital stay, and an inverse correlation between the in-hospital standardised mortality ratio and early post-discharge mortality (defined as 30-day mortality).²¹ This phenomenon was explained by the so called ‘discharge bias’, in which early discharge may be associated with worse outcomes within a short timeframe. It is less likely that such a bias occurred in our study, as we excluded a priori patients with a very poor prognosis.

Finally, as shown in Fig 1, the slope of the Kaplan–Meier survival curve is relatively constant, and no abrupt changes were noticed, especially within the first month after discharge. This certainly is the most important finding, as it confutes the assumption that factors attributable to the hospitalisation per se are preponderant in determining early post-discharge mortality, which instead may be linked to a cumulative effect of dependency and malnutrition, both of which could be minimised with tailored interventions. A study with a prolonged follow-up may be useful for a more in-depth dissection of the slope trend – and its determinants – over time.

Indeed, our study has some limitations. Since the follow-up was conducted with phone interviews, it is possible that certain information was missed. Moreover, our results are limited to an academic, internal medicine setting, and cannot be applied to the primary care setting or a population with a specific medical condition. Also, we did not explore the effects of specific diseases or conditions, as was done in some previous studies. On the other hand, our study also has many strengths, including the wide range of data, that include clinical and performance status parameters, and its monocentric nature. All data were prospectively collected without using administrative data.