Research Critique Part II: The Contribution of Hospital Nursing Leadership Styles to 30-Day Patient, Research Paper Example

Design

The authors are trying to examine a series of hypotheses that will shed light on the following analytical question: Does a correlation exist between the type of nursing leadership style predominant in a hospital and the correspondent 30 day mortality rate? With this research goal in mind, the authors’ decision to employ a secondary data analysis is an appropriate decision. In order to explore this type of question, allocation of a specific intervention via the experimental method would not be viable.  Indeed, the study adopted a non-experimental design primarily because the independent variable –30 day mortality in a hospital- cannot be easily manipulated from a methodological standpoint; however, even if it could be, such an experimental design would not likely be deemed ethical.

The study was cross-sectional in nature; that is, the study analyzed a number of data points extracted from surveys, all at one point in time, based on different leadership styles found in hospitals.  The type of comparison used in the research design is ‘between subjects’ because the subjects are only used once and then not studied again.  There is a lingering question throughout this study of what would constitute a “better” design in order to understand the true relationship between nursing leadership style and 30-day mortality rate in the hospital.  While the use of “between subjects” is proper for an initial study, in order to better understand a possible relationship between the dependent and independent variable, a “within subject” design would be far more illustrative.  This is because, in theory, there may be substantial confounding factors in trying to compare nurses and aggregating it to the hospital level in order to make a comparison.  As a follow-up study, one could try a “within subject” approach for nurses who originally had a low emotional IQ and were given an educational intervention to change their leadership style. Although this study incorporates the use of surveys for secondary data research, this type of research would most accurately be classified as outcomes research.  This is because the study is ultimately looking at how changes in the independent variable affect changes in the outcome variable

There are notable threats to the internal validity of the study.  The first major threat is in how the variable of nursing leadership is operationalized and aggregated to the hospital level.  First, the authors took questions that approximated “emotional intelligence” in order to approximate one of thirteen  emotional leadership competencies that were then aggregated into three different “leadership groups” (resonant-visionary, dissonant-pacesetting, mixed).  The researchers then aggregated that leadership style to define that of each sampled hospital.  There is not only a remaining question of whether the questions themselves are truly representative of “leadership styles”, but also a larger question of whether the nurses sampled reflect the leadership style of the hospital or just one or two isolated care units. This threat to internal validity brings into question whether the study is actually measuring the relationship between nursing leadership style (independent variable) and 30 day mortality (dependent variable).  Because of these reasons, and the general problems associated with drawing causal conclusions from statistical inference, no causal inferences should plausibly be drawn from this introductory study.

There are also questions to what extent the study is externally valid; that is, to what extent can the findings from this study of Alberta hospitals be generalized.   Regarding this point, I would posit that the findings from Alberta would likely be generalizable to similar areas in Canada due to the presence of a national health system that likely has the impact to smooth out nursing quality across the country. In addition, there are both rural and urban hospitals in Alberta which means, at least in theory, both areas may have been sampled to increase the generalizability.  There are definite limitations, however, if one wanted to take the results of this study and apply it outside Canada such to countries as the United States or Singapore where the health care system and health delivery is different than that of Canada.  Overall, the authors are aware of the limited conclusions that can be drawn from a secondary data study that relies on statistical inference.  These concerns, however, were measured against a greater need to test the hypothesis and assess whether there was enough evidence to pursue further research that would use different research designs.

Ethical/ Protection of Human Subjects

There were no pronounced ethical problems in this study, due to the fact that the authors received permission from the health ethics review board at the University of Alberta (Page 333) and due to its non-experimental in nature focusing on secondary data analysis.

Sampling

The sampling methodology of the study is somewhat complex due to the process of aggregation across different sampling units in the study. The authors clearly identify the target population of nurses working at acute hospitals in the Canadian province of Alberta; however, there are some uncertainties regarding how sampling was conducted.  Overall, the authors examine a total of 90 acute care hospitals in Alberta (out of a total of 109) with more than five nurses on staff; the minimum number of nurses was selected in order to provide a more stable data set with less variability (noise).  Once the 90 hospitals were selected, 5,228 surveyed nurses were linked to those hospitals and served as the main sampling unit to determine the “leadership style” of the hospital. Finally, the sample is linked to patients from those hospitals (assumedly) from similar units-although this point is not clear in the paper.  Indeed, the study lacks granular detail in understanding how the sample was ultimately undertaken in any of the major sampling units.  There is no clear reference to the sampling method used among the hospitals chosen besides the initial eligibility criteria.

There are profound questions about whether the nurses selected are representative of the hospitals they are selected to represent. That is, whether the nurses who participated in the original study, and who’s answers were aggregated to represent the hospital do in fact do so.  Overall response rates are not included in the study. No possible sample biases are identified y the authors.  The sample size would appear to be sufficiently powered in this sample in order to detect an effect. The authors did not present a power or other rationale for the sample size.

Data Collection Methods

The paper explains relevant data collection methods.  As mentioned above, there are three main samples found in the data: 1) Hospitals; 2) Nurses; 3) Patients (outcomes).  Regarding the hospitals, the hospitals were selected from data provided from regional health authorities based on institutional characteristics.  Data from nurses was taken from the Alberta Nurse Survey.  Patient data was collected from the Hospital Inpatient Database (Page 333).  There are no explicit discussions regarding data collection methods, the selection of individuals, or regarding the issue of self-reported data in the text.  Fagerstrom comments that the study would be improved by including newer data (the data is 10 years old), as well as including other variables that focus more on comparing nursing work load and intensity rather simply than patient outcomes.

 Validity and Reliability of Instruments

The main issue regarding validity and reliability appeared in the composition of the nursing leadership classification for hospitals.  Indeed, using the nursing predicator at the hospital level introduced issues of validity and reliability (Page 334).  In addition, the authors used the Likert scale to ascertain validity in the survey results gathered from the nurses regarding the hospital leadership environment (Page 334).

Data Analysis

Overall, cardinal and ordinal data (nursing emotional intelligence) was collected to be analyzed.  For continuous variables, means and standard deviations were displayed to describe data; for categorical variables, percentage frequency was displayed.  All the data analysis procedures are clearly described and presented in the article.

The statistical analysis used in the study address the question at hand in a unique way.  That is, because the study dealt with individual-level data (via the Nurses Survey) and aggregate-level data (via the hospitals), the authors decided to use hierarchical logistic regression which specifies random effects at each level of data in order to give a more conservative inference of the overall effect. In addition, the authors built different models in order to answer the different questions contained in the hypotheses: Model 1 was used to answer whether 30-day mortality rates varied across the 90 hospitals sampled; Model 2 was used to test how nursing leadership factors were associated with 30-day mortality rates.  Once the baseline association was established in model 2, the authors then cumulatively added different adjustment factors to examine how much of the initial association could be explained away due to different factors.  Thus, Model 3 adds individual patient factor(s), Model 4 adds institutional factor(s), and Model 5 attempts to examine the overall association between the independent and dependent variable. This statistical modeling did in fact answer the three hypotheses posited on Page 333.

Overall, the paper included a total of four tables: 1) Relationship between Hospital Nursing Leadership Styles (Aggregate) and Individual Patient Factors at the First Level; 2) Relationships between Hospital Nursing Leadership Styles and Hospital Nursing Factors at Level 2; 3) Hierarchical logistic regression analysis results between 30-day Mortality, 4) Individual, and Hospital-Level Variables and Hospital Nursing Leadership Style.  The fourth table is likely the most important table of the paper- it shows the results between nursing leadership styles and 30-day mortality in the adjusted and unadjusted models.  The table information is mentioned in the text without substantial redundant.  The tables’ descriptive statistics, primarily displayed in tables 1 and 2 sufficiently describe the major characteristics of the data set.

The report includes a number of inferential statistical tests: 1) baseline characteristics were examined for continuous variables using Kraskal-Wallis and categorical variables using chi-squared analysis; 2) the main analysis used hierarchical logistic regression. One of the main findings of the study was while there was a significant relationship between nursing leadership style when unadjusted for variables; once the model was adjusted for other predictors, the strength of the relationship attenuated.  Due to the large size of the sample size and representative nature of the hospital chosen, the adjusted results of the study shed some doubt on the researchers overall hypothesis.

Discussion

The discussion section is composed of three main parts: 1) relationship between variables in the study; 2) potential explanations for the relationship; 3) limitations of the study. After the statistical analysis was completed, the adjusted model suggested that resonant leadership style (leadership consistent with greater emotional intelligence) may lower the 30-day mortality rate compared to other models.  In explaining how resonant leadership may lead to a lower overall mortality rate in a hospital, the authors explained that management styles, in addition to nursing attitudes, may play a key role in health outcomes.  In addition, the authors cited evidence that hospitals with resonant leadership had evidenced a higher quality of care through the variable of meeting existing patient needs.  Finally, the authors made the case that homogenous leadership styles, even if they are not inherently resonant, would offer better results than a heterogenous mix of the two that might lead to confusion among hospital personnel.

Finally, the study noted several limitations in their results.  First, the aggregation of individual nursing scores to hospital levels may not be correct- although a validity sub-study showed a significant correspondence between the two. Second, the study was retrospective in nature rather than prospective that constrained the external validity of the study. The authors called for more research to be conducted regarding leadership style on patient outcomes at the regional and national level.

Citations

Cummings, G.G., Midodzi, W.K., Wong, C.A. & Estabrooks, C.A. (2010). The contribution of hospital nursing leadership styles to 30-day patient mortality. Nursing Research. 59(5), 331-339.

Fagerstrom, L. (2011).  The style of nursing leadership in hospitals is an independent risk factor for 30-day mortality of patients in acute care.  Evidence Based Nursing, 14(2), 57.