This article describes the development of a model for risk adjustment of inpatient fall rates in acute care hospitals based on patient-related fall risk factors and presents the impact and results of risk adjustment on hospital performance comparison across Swiss acute care hospitals.
The newly developed risk adjustment model revealed that age, sex, care dependency, fall history, the intake of sedative and or psychotropic medications, surgery and six ICD-10 diagnosis groups are statistically significantly associated with inpatient falls in acute care hospitals in Switzerland. We demonstrated that adjusting for these factors has a relevant impact on the results of hospital performance comparison, as it reduces the number of low as well as high-performing hospitals. In general, it can be stated that the variability of Swiss hospital performance, especially after risk adjustment, was small. Therefore, we can conclude that Swiss hospitals, regardless of hospital type, show a comparable level of care quality with respect to inpatient falls, after adjusting for patient-related fall risk factors.
Inpatient fall risk adjustment model
One of the most crucial steps in the development of a risk adjustment model is the selection of the variables to be used as independent variables in the model. Since we carried out data-driven statistical variable selection in our model development, it is particularly important to critically review the selected risk variables. In accordance with several studies and guidelines [19, 20, 55,56,57,58,59], older age and a fall in the last 12 months proved to be a relevant patient-related fall risk factor in our risk adjustment model. Care dependency also proved to be a relevant risk factor in our model, as well as in the literature [22, 55]. Our study showed that the risk of falling increases with increasing care dependency compared to the reference category “care independent”, with the exception of the category “completely dependent”, which revealed a lower risk of falling compared to the category “to a great extent dependent”, but still a nearly twofold risk of falling compared to the reference category. One possible explanation is that from a certain level of care dependency, mobility is so severely restricted that locomotion is no longer possible or only possible when accompanied by caregivers, and therefore the risk of falling is lower.
The ICD-10 group diagnoses were important to account for relevant comorbidities in the risk adjustment model. The associations between the ICD-10 diagnosis groups selected in the model and the risk of falling in hospital leave room for interpretation. For example, the literature describes that cognitive impairment is associated with a higher risk of falling [19, 20, 22, 55, 59]. Since dementia is classified in the ICD-10 diagnosis group Mental, behavioural and neurodevelopmental disorders, this could be a possible explanation for the selection. This may also be true for the ICD-10 diagnosis group Neoplasms as there is evidence that, in addition to the established general patient-related fall risk factors, cognitive impairment, metastases, especially brain metastases, but also comorbidities such as anaemia or fatigue are specific fall risk factors in cancer care [55, 60]. The remaining ICD-10 diagnosis groups selected into the risk adjustment model seem to be important for hospital comparison but are possibly, with odds ratios between 1.23 and 0.90, of less importance for clinical practice.
Additionally, three statistically significant protective factors, i.e., factors that reduce the risk of an inpatient fall, were also selected into the model. The association between a surgical procedure and a reduced fall risk has also been described by Severo, Kuchenbecker [61]. In contrast, there is controversial evidence on the extent to which the female gender is associated with a reduced risk of falling [20,21,22]. It is also unclear how the ICD-10 diagnosis group diseases of the ear and mastoid process is related to a reduced risk of falling. Especially since a recent retrospective cohort analysis based on a large sample size showed that hearing loss is associated with a higher risk of falling [62]. The result in our study might be related to the relatively small number of patients coded with this diagnosis group. This is also reflected in the relatively wide 95% confidence interval of the odds ratio.
Generally, the intake of sedative and psychotropic medication is described as a relevant patient-related fall risk factor [20, 63, 64]. Nevertheless, it is a moot point whether the consideration of this variable in the risk adjustment model is appropriate due to the procedural character of the variable. Hospitals cannot influence the proportion of patients they care for who have already been prescribed sedative or psychotropic medication, but a rigid prescription regime and medication review on admission might directly influence how many patients receive these drugs during hospitalisation. When deciding whether to adjust for sedatives and or psychotropic medications to increase the fairness of the hospital comparison, the temporal relation of when the medications were prescribed, before or after hospital admission, may be of importance.
This applies in principle to all risk factors in the model. For example, even if it is not possible for a hospital to influence the age of its patients, it can introduce targeted preventive measures for older patients to prevent falls and thus indirectly reduce the risk of falls associated with older age. In this context, the risk model is not only important to enable a fair hospital comparison, but it is also of clinical relevance, as it informs health care professionals which patient groups with which characteristics are particularly at risk of falling. Preventive measures can thus be applied in a more targeted manner. At the process level, the assessment of these factors and the initiation of suitable preventive measures by the nursing staff in daily practice is essential to reducing fall rates in acute care hospital.
While risk adjustment is of central importance in providing a fair external benchmark, risk adjustment may also unintentionally mask potential for quality improvement. For example, a hospital that treats many high-risk patients may be considered to be performing well after risk adjustment, even though the unadjusted inpatient fall rate is higher than in other hospitals. However, this is only the case if the measured fall rate is lower than would have been expected based on the many high-risk patients. Therefore, the respective hospital has already taken preventive measures to keep the inpatient fall rates lower than expected. But in the context of internal quality improvement and the suffering that every single fall means for the patient, the question arises whether it is enough to be as good as the other hospitals. At best, despite the more difficult initial situation with the many high-risk patients, it is possible for this hospital to reduce the inpatient fall rate by further optimising the prevention measures. It may be unfair, but hospitals with many high-risk patients always have to do more to achieve the goal of low inpatient fall rates. Therefore, it might be advisable for hospital management and staff not to look at the risk-adjusted results in isolation, but in combination with descriptive results on patients’ risk factors, preventive measures and effective inpatient fall rates. The overall picture should form the basis for discussion and analysis in the team in order to identify potential quality issues and initiate appropriate preventive measures.
Impact of risk adjustment on hospital comparison
The hospital comparison based on the unadjusted inpatient fall rates revealed 20 low-performing and three high-performing hospitals. In contrast, with the risk-adjusted hospital comparison, it was found that 18 of the 20 hospitals were incorrectly classified as low-performing and that all three of the high-performing hospitals were incorrectly classified. On the other hand, no hospital had been incorrectly classified as an average-performing hospital instead of a low- or high-performing outlier. The study by Danek, Earnest [18], that examined the effect of risk adjustment on the clinical comparison of diabetes-related outcomes showed a comparable effect, as the number of clinics classified as low-performing hospitals decreased significantly after risk adjustment.
In general, the main objective of performance measurements is to provide accurate data to various stakeholders to enable informed decision-making [17]. The non-adjusted hospital comparison as a basis for decision-making would result in some hospitals being ranked better or worse than their actual fall rate performance effectively is. This may have far reaching consequences, especially in health systems where financial reimbursement is directly linked to health outcome measures, as is the case in the US for inpatient falls [65], or if the results are published publicly, which might result in reputation damage for the incorrectly classified low-performing hospitals. In addition to the incorrect classification of low-performing hospitals, our risk adjustment also led to the disappearance of high-performing hospitals. According to Danek, Earnest [18], inaccurate representation of high performance can lead to complacency and have a negative impact on motivation to strive for improvement.
Hospital comparison of inpatient fall rates in Switzerland
The performance of hospitals regarding fall prevention measures is at a comparable level in Switzerland when patient-related fall risk factors are accounted for. The total variance explained by differences between hospitals is 7% in the null model and 3% in the risk-adjusted model. This shows that the variability in performance of Swiss hospitals is generally low and almost disappears after risk adjustment. Therefore, it is questionable if inpatient falls are an appropriate indicator for hospital performance comparison, as only a small amount of variability is explained on hospital level [66]. This is also an ongoing discussion in other research fields such as hospital readmission rates. Hekkert, Kool [67] reported even smaller ICC values of 0.5% to 2.7% at hospital level for readmission rates after different surgical procedures. The National Quality Forum [3] write in their technical report, unfortunately without giving the actual figures, that the ICC of inpatient falls is higher at ward level than at hospital level. Therefore, another question in connection with the low variability between hospitals is whether the wards rather than the hospitals as a grouping variable are of importance. This is supported by evidence that inpatient fall rates vary significantly by ward types. For example, constantly significantly higher fall rates were reported for medical wards than for surgical wards [68]. A risk-adjusted comparison stratified by department type could be a measure to further improve the comparability of the results. However, one problem in examining and comparing ward performance, as in the present study, is that the low number of patients per ward combined with low inpatient fall rates could make the model estimates inaccurate [39]. Continuous measurements with longer survey periods such as monthly, quarterly, or yearly total number of inpatient falls per patient days or the combination of several measurement dates could address this problem.
At the national level, since the variability always refers to the average of all hospitals, no statement can be made as to whether good or bad quality is achieved in Swiss hospitals regarding inpatient falls in general. It is possible that all hospitals perform well or poorly in a homogeneous way. In order to answer this question, risk-adjusted country comparisons, such as the OECD according to Busse, Klazinga [11] is striving for, must be carried out.
Strengths and limitations
Our study is based on a large representative sample, as almost all Swiss acute care hospitals participated in the three measurements. Data pooling of the three measurements increased the number of participants per hospital and protected the hospitals to a certain extent from a random result, which would otherwise have been more likely with a small number of cases at only one measurement point. Still, and unfortunately, some small institutions had to be excluded from the analyses. However, this had the positive effect of creating ideal conditions for the multilevel analyses and thus counteracting possible bias in the analyses. An additional strength of the study was the rigorous, well defined and standardised data collection procedure, which was accompanied by instruction meetings and manuals. This is particularly relevant for hospital comparisons, as another reason for the variation in outcome, besides hospital performance, may be differences in the definition and data collection procedure of inpatient falls in hospitals [42].
One limitation to consider is that our data are based on a cross-sectional design and therefore our findings on the association between fall risk factors and inpatient falls are not causal but correlational. Since the risk adjustment model only considers patient-related fall risk factors, it can be assumed that these factors were already present to a certain extent before the patient was admitted to the hospital (e.g., age, gender, fall in the last 12 months) the significance of the temporal relationship is rather negligible. Nevertheless, care should be taken in further fall measurements to take the temporal relation into account if possible.
In general, it should be noted that a risk adjustment model can only take into account measurable and reportable factors [10, 27]. Other measurable patient-related fall risk factors described in the literature are, e.g., impaired mobility or gait instability [19, 22, 55, 64], urinary incontinence or frequency [22, 55, 61, 64, 69] malnutrition [19, 59] or sarcopenia [19, 70]. In our analysis, however, it was not possible to adjust for these factors as they were not collected in our measurements. The impact of the inclusion of these other factors on the accuracy of the risk adjustment model should be further investigated.
It should be noted that inpatient falls can also be influenced by structural factors at the department level, such as environmental (e.g., floors, lighting [55]) or organizational features (e.g., skill mix, nurse staffing ratio [71, 72]). We did not include these factors in our risk adjustment model because that are exactly the factors which are under the control of the hospital and thus differentiate between hospitals. A risk adjustment for structural factors would limit the incentive for hospitals to review and improve them.