Background
Nursing-sensitive outcomes are “changes in health status upon which nursing care has had a direct influence” [
1,
2] [p 1]. The concept arose from the quality improvement agenda of the Joint Commission on Accreditation of Health Care Organizations (JCAHO) in the United States (US) in the late 1980s [
3]. More recently, concerns about the changing nurse skill mix have led researchers to investigate associations between the incidence of nursing-sensitive outcomes and levels of nurse staffing [
4‐
8].
Based on the literature and expert clinical opinion, Needleman et al. identified 14 potential nursing-sensitive outcomes that could be measured using routinely collected administrative health data. These included: pressure ulcers, deep vein thrombosis and pulmonary embolism, pneumonia, urinary tract infection, central nervous system complications, shock or cardiac arrest, upper gastrointestinal bleed, pulmonary failure, physiologic/metabolic derangement, surgical wound infection, mortality, failure to rescue and length of stay [
9]. Administrative health data are electronic records collected for administrative purposes that include patients’ hospital discharge summaries. These were determined to be the best source for constructing nursing-sensitive outcomes, because they contain diagnoses and procedures coded according to the International Classification of Diseases (ICD) and contain patient level variables such as age, sex, country of birth, and health insurance status in a relatively uniform format [
9]. Needleman et al.’s development of nursing-sensitive outcomes was guided by three criteria: (1) that nursing-sensitive outcomes be conceptually related to nursing care, (2) that outcomes be ‘codable’ from hospital patient discharge (separation) abstracts, and (3) that the outcomes occur in inpatient acute care settings with high enough frequency and variation to allow for statistical analysis [
9] [p 37].
For each outcome, an algorithm (syntax) was developed that used a combination of ICD-9 codes, Diagnosis Related Groups (DRGs), Major Diagnostic Categories (MDCs), length of stay, presence of a surgical procedure and age. Each algorithm included outcome specific inclusion and exclusion (qualifying) criteria in an attempt to include only patients who experienced a truly preventable adverse outcome rather than one associated with the disease process. For example, the ‘pressure ulcer’ algorithm excluded patients with any form of paralysis in their hospital discharge records to ensure those flagged with a pressure ulcer were more likely to have acquired it as a result of the quality of nursing care and it was not present on admission as a consequence of a pre-morbid condition. Algorithms excluded outcomes that were primary diagnoses, and used secondary diagnoses to identify outcomes that were potentially nursing sensitive. As the algorithms developed by Needleman et al. used American ICD-9-CM (Clinical Modification) codes, subsequent translation to ICD-10-AM (Australian Modification) codes using “crosswalks” (mapping keys) was undertaken by McCloskey [
10] in New Zealand. These translated algorithms have been used in Australian studies [
11,
12].
Many of the challenges faced by Needleman and his team [
9] in the matching of data is obviated by the data linkage processes in Western Australia (WA), whereby all patients have a unique identifier that links their individual records into a single ‘chain’ of hospitalisation episodes. Identification of comorbid conditions is enhanced because linked data allows researchers to match records of the same patient both within and across databases, thus providing longitudinal health data on individuals and populations [
13‐
17]. Many patient hospitalisation databases only code patient conditions relevant to a specific episode of care, therefore some chronic conditions may not be recorded in a discharge abstract [
9,
18]. An advantage of using linked data is being able to ‘look back’ to ascertain comorbid conditions for individuals when this is not routinely recorded. Similarly, linked data enables researchers to ‘look forward’ to determine patient outcomes that require hospitalisation and are identified following discharge from the index hospitalisation. The results of a WA study reported that identification of comorbid conditions in an adult population increased from 47% using index hospitalisation data to 90% when a three year look back into administrative health data was undertaken [
17]. Therefore, linked data is likely to provide more accurate information about comorbid conditions which may affect identification of nursing-sensitive outcomes.
The WA Data Linkage Branch adheres to data linkage best practice protocols [
19] and uses probabilistic matching based on medical record number, surname, first name and initial, date of birth, sex and address as the principal variables to link the data. Clerical review of additional information is undertaken for records that fall between definite matches and non-definite matches [
20]. Invalid and missed links have been estimated at 0.11% [
21]. Validation studies have shown recording of additional diagnoses and complications in Health Morbidity Data (HMD) vary between 10-80% sensitivity depending on the nature of the condition [
21].
Studies of nursing-sensitive outcomes in administrative health data have predominantly used adult populations. Some researchers have included children within their adult study populations [
6,
22,
23]; however, few have used administrative health data of paediatric populations [
24,
25]. Given the differences in casemix and risk profiles between paediatric and adult patients [
26,
27], it is questionable whether nursing-sensitive outcomes used with adults are applicable to paediatrics. Optimal paediatric nursing care takes into account the child’s stage of development, consequently age stratification needs to be considered during analysis. Children have fewer chronic conditions and comorbidities and primary healthcare and in-home care delivery are emphasised. As a result, paediatric hospital lengths of stay are shorter than those of adults, which contributes to the challenges of studying paediatric populations. Earlier studies found that there were insufficient numbers of recorded events to make analysis meaningful for some potential nursing-sensitive outcomes [
24,
25]. Many comorbid conditions reported in adults are not present in children so the risks for developing adult nursing-sensitive outcomes are reduced. For the above reasons, it may not be useful to apply algorithms for nursing-sensitive outcomes that were validated in adult populations to paediatric populations. The algorithm for each outcome should be validated within paediatric populations before being used to measure the quality of paediatric nursing care.
In our earlier work with a panel of expert paediatric nurses, we established face and content validity for 17 nursing-sensitive outcomes that were potentially useful for measuring the quality of paediatric nursing using administrative health data [
28]. Seven of these outcomes were also used by Needleman et al., with pressure ulcer, pneumonia and surgical wound infection ranked as top three. The present study focused on these three nursing-sensitive outcomes and aimed to determine whether Needleman et al.’s algorithms, or variations of them, were useful in paediatric populations. A further aim was to determine whether linked data provided more accurate information about comorbid conditions than using data that was not linked, and whether it affected the identification of nursing-sensitive outcomes.
Method
This population based, retrospective cohort study used linked administrative health data from the WA HMD System which enabled ascertainment of all WA hospitalisations for the cohort. The cohort included all WA resident children admitted to one tertiary paediatric hospital during the 10 year period from July 1999 to June 2009 inclusive. Inclusion criteria were that the child had a WA postcode as place of residence, was aged ≤18 years, and had been an inpatient (stayed at least one night; determined by admission date and separation date minus days of care provided by Hospital in the Home). Children who had been transferred from another hospital were excluded and data from subsequent hospitalisation following transfers were not included. Hospital separations are only recorded when the child is discharged or transferred to another hospital. Separation records are not created when the child is transferred within the same hospital. To enable looking back for comorbid conditions and looking forward for possible consequences of the index hospitalisation, HMD from the previous ten years and all subsequent WA hospitalisations were provided for the children in the cohort. The index hospitalisation was the record in which the nursing-sensitive outcome was identified.
The linked HMD provided abstracts of demographic and clinical information on hospital separations from all acute care hospitals within WA. Up to 22 diagnostic variables and 12 procedural variables were provided. Diagnoses were coded using ICD-9-CM until June 1999, and ICD-10-AM since then. Similarly, procedures were coded according to ICD-9-CM until June 1999, but subsequently followed the Australian Classification of Health Interventions (ACHI). Variables for MDC and Australian refined (AR)-DRG were also provided.
Data were received in an anonymised file and analysed in SPSS for Windows (Version 19.0.0.1; 2010 IBM SPSS Chicago, Il, USA), with logical checks undertaken during data clean-up. The translation of ICD, DRG and MDC codes was checked and amendments were made as required. The syntax for each nursing-sensitive outcome algorithm had been written for use in SPSS (Finn, J. unpublished) and this was also checked and amended if necessary. The MDC codes that were used were MDC4 which are diseases and disorders of the respiratory system and MDC9 which are diseases and disorders of the skin, subcutaneous tissue and breast [
18].
Month and year of birth were provided for each child. Age on admission was calculated, then categorised into developmentally appropriate age groups of neonate (1–28 days), infant (29–365 days), toddler (>1-3 years), pre-schooler (>3-6 years), school-age (>6-13 years) and adolescent (>13- ≤ 18 years). The risk pool for surgical wound infections was surgical patients only. This subset of the cohort contained patients classified as ‘surgical’ based on AR-DRG codes which were supplied by WA Data Linkage Branch. The remainder of the cohort was classified as ‘non-surgical’ and contained children coded as medical and ‘other’. [
18]
Process
Each hospital separation record was assumed to be an independent event, therefore calculations were based on hospital separations or records, rather than children, as done in other studies [
7,
11,
24,
29]. Nursing-sensitive outcomes of pressure ulcer, pneumonia and surgical wound infection were identified as per Needleman et al.’s algorithms (Table
1).
Table 1
Definitions of nursing-sensitive outcomes
As per Needleman et al. [ 9] |
Pressure ulcer | Pressure ulcer (ICD2 L89) | All medical and surgical inpatients | MDC9-skin conditions |
| | LOS > =4 days | All diagnosis of hemiplegia, paraplegia, |
| | | paralysis; cerebral palsy; (ICD > =G80 and < =G84) |
Pneumonia | Aspiration, post-operative, hypostatic, bacterial, broncho and unspecified pneumonias: (ICD > =J14. and < =J15.6, J15.8 J15.9 J18.0 J18.2 J18.8 J18.9 J69.0 J95.8, J95.9) | All medical and surgical inpatients | MDC4-respiratory conditions |
| | | Immunocompromised, AIDS; (ICD > =B20. and < =B24., > = D80. and < =D89.9, M35.9) |
All diagnoses of probable community acquired pneumonia (ICD > =J10. and < =J10.8, > = J11. and < =J11.8, =J12. and < =J12.9, J13., J15.7, J16.8, > = J17. and < =J17.8) |
Postoperative pneumonia [ 24] | Aspiration, post-operative, hypostatic, bacterial, viral, broncho pneumonias | All surgical inpatients | |
| (ICD J12. and < =J12.9, J13., > = J14. and < =J15.6, J15.7, J15.8, J15.9, > = J17. and < =J17.8, J18.0, J18.1, J18.2, J18.8, J18.9, J69.0, J95.851, J95.9) | |
Surgical wound infection | Surgical wounds, including surgery post traumatic injury (ICD T79.3, T81.4 T81.41 T81.42) | All surgical inpatients | |
| | Total discharges only |
Recommended paediatric algorithm |
Pressure ulcer | Pressure ulcer (ICD L89) | All medical and surgical inpatients | All diagnosis of hemiplegia, paraplegia, paralysis; cerebral palsy; spina bifida. (ICD > =G80 and < =G84; > = Q05. and < =Q05.9 or > =Q07. and < =Q07.03) |
| | LOS > =3 days | All diagnoses of paralysis found in look back period |
Pneumonia | As per Needleman et al. plus J15.7, J18.1, J69.8 | All medical and surgical inpatients | MDC4-respiratory conditions |
| | | All diagnoses of probable community acquired pneumonia (ICD as per Needleman et al.) |
Aspiration pneumonia and epilepsy (ICD J69.0, 69.8, > = G40. and < =G40.9, > = G41. and < =G41.9) |
All diagnosis of epilepsy found in look back period |
Surgical wound infection | Surgical wounds, including surgery post traumatic injury plus those found in 30 day ’look forward period’ (ICD T79.3, T81.4 T81.41 T81.42) | All surgical inpatients | |
| | Total discharges only |
For each nursing sensitive outcome, frequencies of diagnostic and procedural codes listed in the index records of children with the outcome were calculated to identify any condition or procedure that might have contributed to the outcome, but had not been excluded in Needleman et al.’s algorithm. Selected index records were also examined individually along with previous records (look back) and subsequent separation records of the child to ascertain comorbid conditions which could increase risk. These comorbid conditions were then included as potential qualifying criteria in revised algorithms, which were applied to the paediatric population, and the index records of children determined to have the nursing-sensitive outcomes were identified. For example, to determine the effect of skin condition (MDC9) on the identification of children with nursing-sensitive pressure ulcers, the algorithm was run with and without the exclusion criterion of presence of a skin condition (MDC9).
Each child’s records were aggregated to ascertain the presence of a comorbid condition and, if present, check whether it was noted on the index record. Where necessary, the comorbid condition was added into the index record and algorithms for each nursing-sensitive outcome were re-run using the corrected index records. Based on the results, algorithms were written that should optimise identification of nursing-sensitive outcomes in paediatric populations (Table
1).
If a child had more than one type of nursing-sensitive outcome on the same or separate admissions they were counted as separate index hospitalisations. If a child was readmitted with the same nursing-sensitive outcome, the individual records were viewed. Clinical judgement was used to decide whether the outcome was the same episode or an independent event based on time between discharge and admission, diagnoses and procedures. If judged to be independent events, records were counted as separate index hospitalisations and if judged to be the same episode, then only the first admission was included as the index hospitalisation.
Statistical analysis
Proportions of nursing-sensitive outcomes per 1,000 hospital separations were calculated for the three outcomes. Rates per 10,000 patient days were calculated using Needleman’s algorithms and the revised paediatric algorithms for pressure ulcer and pneumonia, but not surgical wound infection. The definition of surgical wound infection used for both numerator and denominator is ‘within 30 days of surgery’ [
30] [p313]. It cannot be determined when the infection occurred in relation to the surgery from WA administrative health data nor can patients’ lengths of stay post-surgery be ascertained. Therefore, a rate based on total patient days would not provide an accurate measure of the risk pool.
Proportions and/or rates for each nursing-sensitive outcome were determined by running the algorithm with the qualifying criterion included and then rerunning it with the qualifying criterion excluded. For example, the number of children with pressure ulcers who also had a form of paralysis and the number of children with pressure ulcer and no form of paralysis were determined. For each nursing-sensitive outcome a number of qualifying criteria were tested. Estimated rate differences and 95% confidence intervals (CI) were calculated. If there was no statistically significant difference in the rates (CI included zero) or the number of records was <5, the qualifying criterion was dropped from the algorithm. Finally, percentages of records with comorbid conditions found for each nursing-sensitive outcome using look back data and nursing-sensitive outcomes using look forward data were calculated.
Ethical considerations
The project had approval from the Human Research Ethics Committees of the study hospital and the WA Department of Health. To avoid possible identification of children, wherever numbers were less than 5 the exact number is replaced by <5.
Discussion
In contrast to other nursing-sensitive outcome studies using administrative health data [
10‐
12,
24,
32], this study used linked data, which enabled the researchers to identify all WA hospital separations during the previous 10 years for every child in the study cohort. The advantage of this data linkage is that comorbid conditions that are not recorded in the same hospital separation as the one that contained the outcome of interest can be identified. Using the linked data, 29 children with pressure ulcers (43% of all children with a pressure ulcer) were identified as also having a form of paralysis that was not identified in the same hospital separation as the pressure ulcer. Similarly, when using the look back period to identify children who had any forms of epilepsy which increased aspiration pneumonia risk, a further 31 (33%) children with aspiration pneumonia were excluded from the risk set. When records of children identified with these comorbid conditions are excluded from analysis the proportions and incidence rates of the nursing-sensitive outcomes are reduced, and more accurately quantify outcomes that reflect the quality of nursing care. However, the risk of excluding some nursing-sensitive outcomes remains.
The standard for coding additional diagnoses in hospital morbidity data is that the diagnosis must impact on patient care that requires ‘therapeutic treatment, diagnostic procedures or increased clinical care and/or monitoring’ [
18] [p 13]. Older children hospitalised for treatment not related to a comorbid condition, such as spina bifida, cerebral palsy or epilepsy, are often self-caring in relation to their chronic condition; they don’t require an increase in resources, so these conditions are not coded in the index hospital separation abstract. The look back period in this study was 10 years preceding each child’s first separation from the tertiary hospital in the period 1999–2009, which for 77.3% of the children was from birth. Other researchers who have used linked data identified comorbid conditions either in the index separation record or they used a look back period, most frequently just 1 year [
33]. For example, Preen et al. [
17] used regression models to identify the impact of different comorbidity ascertainment look back periods and concluded that shorter periods of look back, approximately 1 year, were appropriate for post-hospitalisation mortality. However, it was suggested that longer look back periods were superior for other outcomes. The researchers reported that of the comorbid conditions recorded 5 years before the index hospitalisation, 46.8% were recorded at the index hospitalisation. This increased to 68.6%, 79.1% and 89.5% at 1, 2 and 3 years of look back respectively. The study was done with adult discharge abstracts and used 102 comorbid conditions identified in the Multipurpose Australian Comorbidity Scoring System (MACSS) [
33]. Further work is required in this area, particularly in the paediatric population.
Using linked data was also beneficial for looking forward to ascertain children’s readmissions to hospitals in WA with a diagnosis of surgical wound infection. When the look forward period was included in the paediatric algorithm (Table
1), the proportion of surgical wound infections nearly doubled. The rationale suggested by Needleman et al. for including surgical wound infection as a nursing-sensitive outcome was related to nurses’ roles in preoperative preparation, which include skin cleaning and antibiotic administration [
9]. However, the nurse’s role also includes postoperative assessment and monitoring, which should lead to early intervention to prevent wound infection. A principle of current paediatric care encourages home care as much as possible and results in shorter lengths of stay following surgery [
34,
35]. As surgical wound infections may not be noticed until a child has been discharged, it is reasonable to consider that children who present with a wound infection within a 30 day period of discharge from a surgical admission have an infection that is potentially nursing sensitive. Occurrence of wound infection could also reflect poor discharge planning with the child and their family regarding post-hospital wound care and medications. If parents are concerned about possible surgical wound infection in their child, children may return to the outpatient department, the surgeon’s private clinic, a general practitioner or emergency department and not require hospital admission with a surgical wound infection. These children are not accounted for in the current data linkage therefore rates are probably underestimating the true occurrence of infection.
Indicators of the quality of care may be used to compare rates of events across time or within and between units and hospitals. We were unable to ascertain whether the additional records with look back and look forward comorbid conditions were randomly distributed between hospitals or over time. If they were, calculations of rates would have consistent errors so including or excluding the extra cases would make little difference on relative performance when used for benchmarking.
As well as amending Needleman et al.’s algorithms to incorporate look back data for certain comorbid conditions, and look forward data for surgical wound infections, the algorithms were further tailored to suit paediatric populations by incorporating variations of qualifying criteria based on their impact in the study population. Differences between proportions or rates of events were analysed with and without each qualifying criterion applied for each nursing-sensitive outcome. For pressure ulcers statistically significant differences were found for each qualifier, except those with MDC9 (diseases of skin and subcutaneous tissues), where numbers were too low for meaningful analysis. Similarly, for pneumonia and surgical wound infection, numbers with immune deficiencies were too low; and for surgical wound infection, numbers with cancer were also too low for meaningful analysis. Although these qualifying criteria have not been included in our paediatric algorithms, regular reviews are recommended as the numbers of children with these diagnoses will alter with changes in treatments, diseases and coding.
Length of stay was analysed to ascertain the most appropriate duration qualifying criterion to apply when determining hospital-acquired pressure ulcers in children. There is a lack of consistency between researchers: some include adult hospital length of stay of longer than 3 days [
9], others use longer than 4 days [
36‐
39]. Curley et al. [
40] found evidence of hospital acquired pressure ulcers in children as early as the second day of their hospital stay when assessing children in intensive care units during a point prevalence study. Number of pressure ulcers was low in our study when the length of stay criterion was less than 2 days, versus 2 or more days, so the statistical significance of the differences in rates using this criterion should be interpreted with caution. However, it is recommended that the paediatric algorithm for pressure ulcer include children who had a length of stay of longer than two days.
Patients with cancer are often immunocompromised and at increased risk of developing infections. Therefore, algorithms for nursing-sensitive outcomes that are the result of infections often include a qualifying criterion that removes this high risk group. The difference in rates of pneumonia in children with and without cancers was not statistically significant. As the rate was higher in children without cancer than those with cancer, this qualifying criterion was not included in the paediatric algorithm. Needleman et al. also removed it from their algorithm as it did not enhance the specificity of identifying outcomes in adult populations [
9]. On the other hand, the criterion has been retained by others when calculating incidence of outcomes that are hospital acquired infections based on clinical reasoning [
38,
39]. The latest version of outcomes recommended by the Agency for Healthcare Research Quality (AHRQ) has not excluded children with cancers but stratified them into different risk groups [
41]. In larger populations the number of children with cancer and pneumonia may be greater and stratification or exclusion of this group of children may be necessary.
As pneumonia is frequently used to indicate the quality of nursing care, [
7,
24,
42] and the results of our earlier Delphi study [
28] prioritised subgroups of pneumonia as nursing-sensitive outcomes, proportions and rates of three subgroups were calculated. The ICD codes for the individual pneumonia types were included in Needleman et al.’s algorithm. Ventilator associated pneumonia, which is particularly used to indicate the quality of care in intensive care settings, had no records coded despite children in the cohort requiring 381,316 hours of ventilation. Follow up with the coders confirmed that there were no records coded as they did not identify any diagnoses of ventilator associated pneumonia in the medical records (Logan, J. personal communication). Postoperative pneumonia, which is applicable to surgical patients only [
24], appears to be a viable indicator of the quality of nursing care, but there were too few records of aspiration pneumonia in our cohort to evaluate it.
This cohort of hospitalised children is similar to cohorts of children in public hospitals across Australia. More boys than girls under 15 years of age are hospitalised [
43] and the most common reasons for hospital admission are respiratory and gastrointestinal conditions in children aged less than 14 years, and rates of injury, poisoning and other external causes increase with age [
44,
45]. Other leading causes of hospitalisation are chronic diseases of tonsils and adenoids [
45].
Limitations
A limitation of using WA administrative health data for identifying nursing sensitive outcomes is the inability to determine whether secondary diagnoses are pre-existing comorbid conditions or complications that occurred during hospitalisation. The use of linked data can reduce this limitation by identifying children who have comorbid conditions recorded in previous separation records and can be excluded from being considered to have a nursing-sensitive outcome.
A further limitation is that the date of surgery was not included in HMD, so the 30 days post-surgery timeframe assumed that surgery occurred on day one of the hospitalisation. There is potential to have under estimated the rate of surgical wound infections as patients who had their surgery later than day one would not have had 30 days post–surgery included in the analysis. However, it may be more accurate than no outer time limit. [
46,
47] To increase the accuracy of this indicator, and others which measure quality postoperatively, it would be beneficial if the date of surgery became a routine variable in HMD.
There is a likelihood of under or over reporting of outcomes. Under reporting occurs particularly when the outcomes have no financial implication for the health service provider [
9,
48]. Not all nursing care is recorded in patient’s clinical records [
25,
27] and not all of the clinical records are coded into the separation records [
18]. However, an audit of coding of ICD-10-AM showed a high level of reliability and adherence to coding standards [
49]. Validation studies using review of hospital charts have found a reasonably high level of data accuracy and reported 87% accuracy for DRG coding within WA HMD [
21]. The advantages of feasibility, cost saving and having complete longitudinal population data when using administrative health data offsets the limitations and can provide reliable population-based estimates of nursing-sensitive outcomes.
As numbers of nursing–sensitive outcomes are small in this WA paediatric population, comparisons of differences in rates may not be reliable, particularly when there are fewer than 5 records [
50]. It is important to consider the number of records when interpreting results.
Implications for research and policy
Using linked data is advantageous for identifying comorbidities that are not recorded on index separation records, but exclude children from being considered to have nursing-sensitive outcomes. When comorbidity records from previous separations are included, the specificity of the algorithm is increased, and a more accurate number of actual nursing-sensitive outcomes can be ascertained. Furthermore, linked data assists in identifying outcomes that become evident following discharge. When using administrative heath data, linked data should be used, particularly if all comorbid conditions are not routinely identified in the index separation record and to assist in the accurate identification of outcomes.
Numbers of pressure ulcers recorded in administrative health data in children are too few to be a useful measure of the quality of nursing care in this population. This finding is similar to results of a study undertaken in a large paediatric cohort in California [
24]. However, the algorithms for pneumonia and surgical wound infections could potentially be considered for quality improvement initiatives within a hospital. As the health system and paediatric cohorts are similar across Australia [
51], these algorithms are a starting point for national benchmarking to compare events and rates within tertiary paediatric hospitals. However, before being used for benchmarking between hospitals or areas within a hospital, additional risk factors must be identified to allow appropriate stratified analysis. Furthermore, paediatric algorithms can be used to determine whether there are associations between nursing-sensitive outcomes and levels of nurse staffing that would assist in ascertaining appropriate staffing levels in paediatric hospitals.
Competing interests
The authors declare that they have no competing interests.
Authors’ contribution
SW conceived the study, secured funding, carried out the study design, data analyses, drafted the manuscript. AB assisted with data analyses, revised manuscript. YH revised manuscript. JF conceived the study, secured funding, assisted with data analyses, revised manuscript. All authors read and approved the final manuscript.