Effects of Patient Volume, Staffing, and Workload on ICU Outcomes 

Abstract & Commentary

The UK Neonatal Staffing Study Group performed a prospective evaluation of a variety of ICU structure measures and their relationships with ICU outcomes. Tucker sought to determine which structure variables, mainly patient volume (workload), nursing provision and consultant staffing, were closely related to the outcomes, namely mortality, mortality or cerebral damage, and nosocomial bacteremia. The study involved neonatal ICUs (NICUs). From a total of 186 NICUs in the United Kingdom, Tucker invited 54 randomly selected ICUs to participate in the study. They had selected these ICUs to represent a balanced 3 × 2 × 2 design. The NICUs were classified according to number of low birth weight infants (< 1500 g) admitted per year in high, medium, and low volumes. The NICUs were further subclassified based on nursing provision (high, with nursing provision 0.84 times or higher than the national minimum standard, vs lower than this) and consultant provision (high and low, with high being more than 1 neonatal specialists per unit and low being one neonatal consultant only). Thus each NICU was stratified and assigned to one of the 12 groups.

Three outcomes were of primary interest. These included hospital mortality, hospital mortality or major cerebral abnormality of probable postnatal origin (cystic leukomalacia or porencephalic cyst on cranial ultrasound arising more than 10 days after birth), and probable nosocomial bacteremia (first positive blood culture more than 48 h after birth). Attributable hospital deaths excluded lethal congenital abnormalities, deaths after complex cardiac surgery or organ transplantation, and resuscitated stillbirths. Deaths were attributed to the hospital of care only if the infant was transferred to that hospital before 24 h of age, or transferred out of it after that age.

Tucker used 2 sets of physiologic variables to adjust the patientdependant variables. One score based on measurements at birth and another score based on measurements made 12 hours after birth were used to adjust for the differing risks that the neonates may have had for an adverse outcome. Using multiple logistic regression and other techniques to confirm that the models of risk adjustments were functioning well, they developed various models to test their hypothesis after risk adjustment.

During the study period, 14,611 consecutive infants were admitted to these 54 NICUs. Information was available on 14,343 infants, of whom 13,515 were eligible for the study and 99% of them had complete data. There were 393 deaths in the hospital, giving a crude mortality rate of about 3%. More than 80% of deaths were attributable to participating hospitals. The unadjusted odds of all outcomes were lower in low volume units.

The unadjusted odds of development of nosocomial bacteremia were lower in low-volume units than in high-volume units and in low consultant availability units than in high-consultant availability units. With the birth or 12 h model, there were no differences in the odds of any primary outcomes by study organizational characteristics (patient volume, nursing provision, or consultant provision) with the exception of nosocomial bacteremia in relation to consultant provision. After adjusting for the risk factors, the odds of developing nosocomial bacteremia were higher in NICUs with high-consultant availability. Also, the odds of mortality seemed to rise as occupancy rose and as nurse-to-infant ratio rose. An analysis of mental health of the staff, assessed by a self-reported questionnaire, showed that less than 2% of staff reported mental health problems (Tucker J. Patient volume, staffing, and workload in relation to risk-adjusted outcomes in a random stratified sample of UK neonatal intensive care units: A prospective evaluation. Lancet. 2002;359[9301]:99-107).

Comment by Uday B. Nanavaty, MD

Assessment of large health care systems has become increasingly sophisticated with the advent of current technology. It has become increasingly common for ICUs to gather large amounts of data, including physiologic or clinical data, in one common platform. Such data are then combined for several such ICUs to get an assessment of large health care systems such as in a region, a state, or perhaps, as in this example, across a whole country. The study presented here suggests that the outcomes in NICUs do not really depend upon such ICU structural measures as volume of at-risk neonates or nursing or physician staffing. It suggests that workload (i.e., how much of the NICU is occupied and how many patients the nurses are actually caring for) may be a more important determinant of such outcomes as mortality or mortality and cerebral damage.

After analysis of nearly a third of the NICUs in the UK for more than a year, this study seems to give a clean bill of health for small NICUs. In the context of drives to centralize services under one huge umbrella, or to consolidate smaller hospitals into one giant medical center, this study would argue that in fact these smaller NICUs may be performing at par and probably contribute greatly to serving small rural communities. In fact, this study actually seems to be raising a red flag for large hospitals with multiple consultants, as the rate of nosocomial bacteremia was higher for high consultant hospitals after adjusting for risk factors.

From reading the paper and all the details of the wealth of data with it, one complex question remains unanswered. Low volume NICUs tended not to have taken patients that required complex cardiac surgery or organ transplantation. The high volume NICUs took care of these patients, and their deaths were excluded from analysis. How should we assess the effects of removing such patient from analysis, with respect to the effect of staffing and mortality? Does excluding them from analysis make these NICUs seem to do better by keeping their census up but lowering the actual mortality? If these deaths were excluded from analysis, did they contribute to calculation of staff to patient ratios or occupancy rates?

Quality of care is an age-old debate in medicine. Florence Nightingale may have started some of the debate when she suggested nearly 150 years ago to improve sanitation in hospitals to reduce mortality and improve outcome. It was a routine practice in England in the late 1800s to collect and publish mortality statistics. In the United States, the debate about quality of care grew stronger in the 1980s and currently it is raging as researchers, policymakers, providers, and payers try to figure out the best ways and means of evaluating quality of care. Clinicians always take pride in providing "highest quality care." But it is hard to define what constitutes quality of care. Following Donabedian’s models, quality has been assessed in terms of structure, process, and outcome. It is well known now in adult intensive care literature that there is something like a "good death." It may be that not the ends (mortality in this case) but the means to the ends (the process of taking care—what happens at bedside) should be studied in greater detail.

Unfortunately, process—what physicians and other providers do—is a hard thing to measure. Nonetheless, to a provider, that is the most important thing to measure. Because process is hard to measure, researchers have examined structure and outcome and derived associations between them. That may also be a valid approach for policy makers. For providers and for educators of the providers, more research needs to be done in the process part of care. What we do and how we do it at the bedside will be the next phase of quality of care research. Hopefully, it will result in improved outcomes and satisfaction with the measure from the providers’ perspective.