The importance of the human resources management (HRM) to the success or failure of health system performance has, until recently, been generally overlooked. Health sector reform in many countries in the 1990s focused on structural change, cost containment, the introduction of market mechanisms and consumer choice [1, 2] but with little direct attempt to address HR aspects.

In recent years it has been increasingly recognized that getting HR policy and management "right" has to be at the core of any sustainable solution to health system performance [3, 4]. This is partly a result of the need to upscale capacity in many country health systems to meet the Millennium Development Goals. A well-motivated and appropriately skilled and deployed workforce is crucial to the success of health system delivery. The actual methods used to manage human resources in health care may in themselves be a major constraint or facilitator in achieving the objectives of health sector reform [5, 6].

In comparison to the evidence base on health care reform-related issues of health system finance and appropriate purchaser/provider incentive structures, there is very limited information on the HRM dimension or its impact [7]. There is a limited, but growing, evidence base on the impact of HRM on organisational performance in other sectors, but there have been relatively few attempts to assess the implications of this evidence for the health sector. This paper examines this broader evidence base on HRM in other sectors and examines some of the underlying issues related to "good" HRM in the health sector.

HRM in health has to function in a sector with some unique characteristics. The workforce is large, diverse, and comprises separate occupations often represented by powerful professional associations or trade unions. Some have sector-specific skills; other can readily move from the health sector to employment in other sectors. The avowed first loyalty of those with sector-specific skills and qualifications (physicians, nurses, etc.) tends to be to their profession and their patients rather than to their employer.

In many countries, access to health professional training and employment is controlled by standards and entry requirements determined by the professions, and aspects of their work are regulated. The health sector is a major recipient of public and/or private expenditure, and health care delivery is a politicised process.

Whilst many health systems have been attempting to decentralise to improve efficiency, they tend to be characterized by a broad range of active stakeholders, a high level of direct and indirect governmental and regulatory intervention, and recurrent "top-down" attempts at reform. Health is also very labour-intensive – the proportion of the total spent on staff is much higher in health than in most manufacturing industries and in many service industries.

The irony is that whilst HR is under-researched in health, partly because of its unique context, the main "business" of health – clinical interventions – is the subject of continuous and detailed research-based scrutiny. No other sector has the same level of self-critical focus, with the use of sophisticated methods such as randomised control trials (RCTs), systematic reviews and meta-analysis.

Research on human resources in health, as in any sector, is drawn from a broad range of disciplines, (e.g. economics, econometrics, occupational psychology, sociology). Some of these disciplines are not open to the use of the types of methods favoured in clinical research in health care. The challenge for researchers attempting to build the evidence base on HRM in the health sector is that they have to draw on these non-clinical research methods to assess the HRM "inputs" whilst attempting to identify appropriate and sector-specific measures of process, output or outcome.

There have been a few attempts to examine "high performance" HRM characteristics in the health sector. These tend to be based in North America, which may be partly, at least, a reflection of the greater focus on "business" practice in private-sector health care industries in the USA and Canada.

Eaton [18] examined issues related to "high performance" HRM in 20 nursing homes in the USA. She reported that quality outcomes at some of the homes improved after reorganisation that included implementation of a new model of HRM based on job enlargement and cross-training, but concluded that the "business" focus of the "high performance model" made it inappropriate for the health care sector. Rondeau and Wagner [19] examined the impact of HRM practices and the contingency theory on 283 Canadian nursing homes. They reported that the "best performing" nursing homes (as measured by indicators of client and staff satisfaction, operating efficiency and revenue) were found to be more likely to have implemented "progressive/high performance" HRM practices and to have a workplace climate that strongly values employee participation.

Studies that can access "business" performance data in health are few in number. There is a broader and deeper evidence base that focuses on health sector-specific measures of process, activity or outcome, and attempts to link these to HRM interventions.

One area where there has been a significant growth in research has been large-scale studies examining links between staffing levels, mix and outcome in the last two years. Whilst not directly addressing specific HRM interventions, these studies make two contributions: they add to our understanding of the linkages between staffing and outcomes, and they also provide a test bed for identifying and assessing the appropriateness of outcome indicators in relation to staffing. Recent studies include:

Most of these studies have reported that higher staffing levels and/or staffing mix are related to "better" outcomes, however defined – either to reductions in the levels of specified "negative" outcomes (such as mortality rates, needlestick injuries, or infections) or improvements in reported quality of care or patient satisfaction.

Whilst most of these studies are "one-off" examinations, there is one series of loosely-linked studies in the USA that has investigated the so called "magnet hospitals". These institutions have been identified as being successful in recruiting, retaining and motivating nursing staff. Research on magnet institutions has been under way for over 20 years, and has highlighted positive links between good human resource practice, staffing characteristics and outcomes of care.

The genesis of the research on magnet hospitals was the policy study published in 1983 [30]. The study identified the organisational characteristics of hospitals that "serve as magnets for professional nurses: that is, they are able to attract and retain a staff of well-qualified nurses and are therefore consistently able to provide quality care. The term "magnet" was used to highlight the staff attraction/retention characteristics of these institutions.

The key characteristics of magnet hospitals, as identified in this first report, included participatory and supportive management style, decentralised organisational structure, clinical career opportunities, planned orientation of staff, and an emphasis on in-service/continuing education.

The report stimulated other researchers to focus on organisational attributes, HR practices and measures of HR "success". There has since been a series of research studies on aspects of "magnetism". These have included studies reporting that nurse turnover and vacancy rates in the magnet hospitals were significantly lower, and reported nurse job satisfaction higher, than in the comparator hospitals [31] and a study examining mortality rates in 39 magnet hospitals and 195 control hospitals using multivariate matched sampling to control for hospital characteristics. The study found that magnet hospitals had a 4.6% lower mortality rate for Medicare patients than the control hospitals [32].

The evidence base on "magnet" organisations has grown and broadened over the two decades since the original study. The main message from the various studies is that "magnetism" does appear to be related to "better" staffing indicators, and to improved quality of care. This has been attributed by Aiken and others to the sustained implementation of a "bundle" of HRM interventions that fit with organisational priorities and that support autonomous working by nurses, enable participation in decision making, facilitate career development and enable high level skills to be deployed effectively.

Some of the indicators, particularly the clinical indicators, are likely to be routinely reported only in health systems with a relatively sophisticated information infrastructure. And some of the indicators (e.g. patient length of stay) may be proxy measures for costs. It should also be noted that there is some evidence that not all the outcome-specific indicators in the final section of Table 2 are universally applicable. The large-scale study of nurse staffing and outcomes in the USA [22] considered and rejected some indicators and reported that some outcome indicators are more sensitive than others in particular types of care delivery.

It is clear that it is not only the organisational context that differentiates the health sector from many other sectors, in terms of HRM. Many of the measures of organisational performance in health are also unique. As noted in the previous section, "performance" in the health sector can be fully assessed only with indicators that are sector-specific. These can focus on measures of clinical activity or workload (e.g. staff per occupied bed, or patient acuity measures), on measures of output (e.g. number of patients treated) or, less frequently, on measures of outcome (e.g. mortality rates; rate of post-surgery complications). The challenge for researchers and policy analysts in the health sector is to bridge the current knowledge gap – between what we know from the general evidence base on HRM inputs and performance, and what we know from the health-specific evidence base focusing on sector-specific outcome measures.

The other main lessons from the evidence base examined in this paper are the need to consider "contingency" – that there must be a "fit" between the HRM approach and the characteristics, context and priorities of the organisation in which it is being applied; and the recognition that so-called "bundles" of linked and coordinated HRM interventions will be more likely to achieve sustained improvements in organisational performance than single or uncoordinated interventions. In the often "politicized" health sector, this is an important message.

Finally, it should be noted that defining the "best practice" evidence base is one thing, but translating this into widespread application of the appropriate bundle of HRM interventions is another. Both Richardson and Thompson [10] and Guest [16] highlighted the issue of the relative lack of "take up" of HRM good practice: even when it has been verified by the studies quoted above, it is not evident in day-to-day practice in many organisations. This highlights an important issue for any sector wishing to improve HRM practice: deciding how best to disseminate good practice in HRM is as important as determining how to identify and evaluate it.