This longitudinal observational study explored the association between nurse staffing levels and adverse mortality on medical and surgical wards in Belgian hospitals. We found that wards with on average higher mean NHPPD’s had lower mean proportions of patients experiencing unexpected death, CPR and death after CPR. Furthermore, we estimated the adverse mortality by combining the mortality after CPR and after unplanned ICU admission with the unexpected death rate and found a significant negative relation between NHPPD and the combined mortality rate. These analyses included adjustment for the unbalanced data between wards, across time, and for important clinical confounders such as nurse’s education, nurse’s support staff HPPD, patient’s comorbidity and mean age. The most apparent limitation of this study was that it included a post-hoc analysis of data collected for another study investigating the effect of a Rapid Response System on patient outcomes. However, in this previous study we purposefully collected data on nurse staffing levels and patient morbidity because we hypothesised, considering the existing literature, that they could be important confounders when studying patient mortality. Moreover, we only used control group data of the previous study to avoid interference of the intervention on patient outcomes. Daily data concerning staffing and patient comorbidity were not available. Therefore, we aggregated data to the ward level resulting in staffing levels and patient comorbidity estimates per period. Since the staffing levels and patients vary from day to day on a specific ward, and because of the relatively low incidence of unexpected death, it seemed sensible to calculate estimates from partially available data collected during a long period of time. In future research however, it is advisable to collect data concerning all admitted patients and staffing levels per admission day to improve the precision of ward level data. Another limitation of this study was that we missed patients’ comorbidity scores from one hospital at two different time points because of technical difficulties in obtaining this data. Moreover, a comorbidity score does not account for complicated procedures or acute changes in the patient’s condition. Additionally, we did not include patient turnover as a possible confounder in our analysis. Therefore, we may have underestimated nurses’ experienced workload and its possible effect on outcome [
20]. We did not collect ICU characteristics such as ICU medical-nursing staffing levels or ICU bed availability which could have influenced the composite mortality outcome. Since we only included Belgian hospital wards in this study, the generalisability of our results is limited. Our study is one of the few studies investigating the effect of nurse staffing levels on adverse mortality on medical and surgical wards using ward level data instead of aggregated hospital-level data [
1,
8]. Evidence suggests that nurse-to-patient ratios and nurse education both influence many patient outcomes including in-hospital mortality. However, crude patient mortality has been criticized for its lack of sensitivity to nursing care [
21]. This could be the case because of the possible low impact nurses have on the total (crude) mortality since some patients’ deaths are inevitable. In our previous study we showed that an important part of the crude mortality rate were patients who died after receiving a do not attempt resuscitation (DNAR) code [
14]. Moreover, when we applied a new definition for unexpected death, we discovered that its incidence is much lower than anticipated and it does not correspond to the crude mortality rate. In this study we used the same outcome indicators and found that nurse staffing was not associated with the crude mortality rate, but it does have an impact on unexpected death, death after CPR and the combined mortality rate. Patient surveillance is one of the nursing activities that are frequently neglected when workload is high [
5,
22]. Missed patient deterioration could result in cardiac arrest with CPR or even unexpected death. In this study we used the combined mortality rate as an estimate for adverse mortality which included unexpected death, death after CPR, and death after an unplanned admission to the ICU. This measure comprises adverse deaths and excludes the expected mortality on the general ward (i.e., death of a patient with an untreatable disease). Moreover, a patient’s death after CPR or transfer to the ICU could be the consequence of late detection of patient deterioration (missed care). Therefore, we think it is much more sensible to use the combined mortality rate, estimating adverse death, when studying the effect of nurse staffing levels. Although this was an observational study, we used longitudinal data which enabled us to take into account the interference of time. Moreover, we adjusted for clustering and other clinical confounders. This makes a causal interpretation much more plausible but certainly not definite. Because of the limited amount of multicentre studies collecting longitudinal data at the ward level, it is very difficult to transfer research results into practice [
10]. In a recent study, Fagerström and colleagues collected data from 36 wards in four Finnish hospitals on a daily basis and demonstrated a relation between nurse workload and patient mortality [
23]. However, they did not include the nurse’s education level or the potential influence of other team members supporting patient care. Another recent longitudinal study by Griffiths et al. (UK), also found a significant relation between low nurse staffing levels and an increased risk of death [
24]. The authors of this last study included registered nurses and unregistered nursing assistant staffing levels, but they only collected data in one acute care hospital. To our knowledge, our study is therefore the first longitudinal, multicentre study including nurse staffing levels, nurse education and care team composition with data collected on the ward level. In a key European study from 2014, Aiken and colleagues showed that every 10% increase in the number of bachelor’s degree nurses was associated with a decreased likelihood of death by 7% within 30 days of admission [
3]. The authors also concluded that increased nursing workload was correlated with patient mortality. We found a mean NHPPD of 2.48, consistent with the known numbers for Belgium (24 h / 10.8 patients per nurse = approximately 2.22 NHPPD) [
25]. Furthermore, the proportion of nurses with a bachelor’s degree was also comparable to previous research (0.59 in this study vs. 0.55 by Aiken et al.). The mean crude mortality rate in our study was, however, much larger (2.05% in this study vs. 1.20% by Aiken et al.). The most evident explanation for this difference is that Aiken et al. only included surgical patients, which may result in lower comorbidity scores and a reduced mortality rate [
26]. We confirm previous results concerning the effect of nurse staffing and nurse education level on patient mortality. Importantly, the composition of care teams is heterogeneous around the world and even across Europe. Even nursing degree levels are not uniform in the European Union [
27]. Therefore, investigating the impact of the care team composition on patient outcomes in an international context could be challenging. It seems evident, considering our results and previous research, that the proportion of highly educated nurses has a significant impact on patient safety and subsequently on mortality. Hospitals and even governments could calculate the optimal proportion of nurses with a bachelor’s degree per ward using the same method as described in this study. Furthermore, the Institute Of Medicine released a report recommending that the proportion of nurses with a bachelor’s degree should be 80% by 2020 to provide safe care [
28]. Our figures indicate that this is at the moment certainly not the case in Belgian hospitals (mean proportion of 59% bachelor’s degree nurses). During the last decade, researchers have been studying care processes that may prevent patients from experiencing harm during their hospital stay [
29]. Since nurses have a very important role in detecting deteriorating patients on the general ward, many interventions were therefore targeted at nurses. However, the increasing care demands in acute care hospitals, combined with the nursing shortage could impact the effectiveness of such interventions [
30]. Policy makers should be aware of this ongoing issue and adequate staffing levels should be determined to provide safe care [
31]. Ward based minimal staffing levels should be considered relating to ward acuity measures estimating nurse’s workload and patient outcomes. Using the regression equation resulting from our analysis, we attempted to calculate the theoretically optimal NHPPD per ward. Interestingly, this allowed us to discriminate between low acuity and high acuity wards since it takes into account patient comorbidity, age and it also allows adjustment for team composition. This method could be used to calculate and adjust nurse staffing levels to a changing care environment. We compared our results with the mandatory NHPPD provided by the Australian government to provide safe care [
9]. The actual mean NHPPD in Belgium is lower than the minimal mandatory NHPPD for hospital wards provided by the Australian government (1.75 in this study vs. minimal 3.00 for one-day hospitalisation including day surgery and dialysis). Furthermore, the highest optimal NHPPD calculated in this study is still lower than the minimal mandatory NHPPD for high complexity acute care wards in Australia (5.36 in this study vs. minimal 5.75 for category C non-ICU wards). This shows that Belgian acute care hospitals have comparatively low nurse staffing levels and that the calculated optimal NHPPD corresponds roughly to the Australian mandatory staffing levels.