Methods
The literature review was based on two sources of information: online databases and reports from governments and professional organizations. The databases were searched for English-language articles on performance measurement published between 1995 and July 2002 and included MEDLINE, PsycINFO, PubMed, CINAHL, and HealthSTAR. The following phrases were independently used in the search: performance measure, quality indicator, process measurement, outcome measurement, and quality of care. The search was focused on measures used in health and mental health care. In addition, a Web search of online government reports and professional practice organization reports was performed. Citations in articles were reviewed, and advice was sought from experts in the field to identify additional performance measures.
The database searches yielded a total of 492 unduplicated references, with appropriateness and eligibility for inclusion in the current review determined by abstract screening. Inclusion criteria consisted of the following: the focus of the abstract was performance measurement or quality of care evaluation and either the abstract represented a review of performance measure work or the abstract presented research evidence that was based on at least one identified measure with face validity. A total of 142 references met the inclusion criteria and were individually reviewed. The overall distribution of publications by country was as follows: the United States (95 publications, or 67 percent), the United Kingdom (27 publications, or 19 percent), Canada (17 publications, or 12 percent), and Australia (three publications, or 2 percent). In total, 73 performance measures were identified in the literature, including eight with categorical definitions. These measures were classified into eight domains defined by the Canadian Institute of Health Information (
19). The domains included acceptability, accessibility, appropriateness, competence, continuity, effectiveness, efficiency, and safety.
Professional and accrediting organizations have published guidelines and standards, and these were also included in our list of performance measures (
20,
21,
22,
23). In addition, a number of government-initiated or government-sponsored reports and references were identified. The U.S. National Inventory of Mental Health Quality Measures was developed by the Center for Quality Assessment and Improvement in Mental Health. The inventory is a catalogue of measures that are operationalized, evidence based, and empirically tested. It is available for use at www.cqaimh.org.
In the United States Hermann and colleagues (
24) conducted a review of measures proposed for application to mental health care that were specific to schizophrenia. Forty-two process measures were identified. Twenty-five measures (60 percent) were based on research evidence that linked measure conformance with improved patient outcomes. Only 12 measures (29 percent) were fully operationalized. Few were tested for reliability or validity. The authors aptly state that these data provide a "snapshot of the status of schizophrenia process measurement amid its ongoing development."
In Australia a set of performance measures was developed to monitor the progress of the National Mental Health Strategy (
25); also available is the Australian Clinical Guidelines for Early Psychosis (
26).
A key innovation in the United Kingdom is the development of National Service Frameworks, which intends to set national standards and define service models for specific services or care groups, to design programs that support implementation, and to establish performance measures for use in creating benchmarks (
27). The National Center for Health Outcomes Development recommended a set of 20 outcome measures for severe mental illness (
28).
A number of performance measures that assessed clinical status (effectiveness domain) were found in at least two types of sources—that is, government reports, published literature, and professional practice organization reports. In addition, most of the measures within the acceptability and appropriateness domains were similarly found in at least two types of sources (78 percent and 65 percent, respectively). This finding would suggest that our searches had reached a degree of saturation.
Table 1 provides the descriptions and the sources (
19,
20,
21,
22,
2326,
27,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38,
39,
40,
41,
42,
43,
44,
45,
46,
47,
48,
49,
50,
51,
52) of the performance measures in all eight domains.
The Delphi technique has been widely used in health care research and mental health services research, including in the identification of key components of schizophrenia care (
53), the description of service models of community mental health practice (
54), the characterization of relapse in schizophrenia (
55), and the identification of a set of quality indicators for primary care mental health services (
56).
Although historically the Delphi technique, a consensus method, has been used with a panel of experts, it has been argued that it is important to broaden the stakeholder groups to include clinicians, consumers, payers, and providers (
10,
57). Although experts in developing and evaluating the evidence base need to be involved in selecting performance measures, it is vital that the perspective of other stakeholders be included. For example, the organization that pays for the service will likely focus on the cost-effectiveness of care, whereas consumers will likely focus on access and acceptability. Another benefit of including multiple stakeholders is to garner support for the implementation of the services (
58). In addition, the consensus technique was used to reduce the number of measures to a more manageable amount.
Consensus methods are structured facilitation techniques that explore consensus among a group by synthesizing opinions. Although a variety of consensus techniques exist (
59), all sharing the common objective of synthesizing judgments when a state of uncertainty exists, the Delphi has four important features. First, it is characterized by its anonymity, thus encouraging honest opinion free from group pressure (
60). This method is an advantage when both consumers and clinical experts are included, lest the experts dominate discussions. Second, iteration allows stakeholders to change their opinions in subsequent rounds. Third, controlled feedback illustrates the distribution of the group's response, in addition to the individual's previous response. Finally the Delphi technique can be used to engage participants who are separated by large distances because it can be distributed by mail or online (
61). This method therefore was appropriate to use in the selection of a core set of performance measures for application to early psychosis treatment services.
The list of performance measures that resulted from the literature review was developed into a Delphi questionnaire. This questionnaire was first pilot tested and refined with three individuals who were familiar with early psychosis treatment services. Pilot testing involved individual, in-person administration of the questionnaire by a research coordinator to a patient, family member, and staff member in the local early treatment program for psychosis. The research coordinator asked these three individuals about the clarity of the instructions, definitions, and descriptions of the performance measures.
The following eight domains constitute the framework: acceptability, accessibility, appropriateness, competence, continuity, effectiveness, efficiency, and safety. Definitions were provided for each domain and for each of the five ratings on a Likert scale. Possible scores ranged from 1 to 5, with 1 representing essential and 5 unimportant.
In the next phase of the study, questionnaires were presented in three rounds to a panel of purposefully selected stakeholders. Purposive sampling is a nonprobability sampling technique in which participants are not randomly selected but instead are deliberately selected to capture a range of specified group characteristics. This form of sampling is based on the assumption that the researcher's knowledge of the population can be used to carefully select individuals to be included in the sample (
62). For this particular study purposive sampling is superior to the alternatives because the stakeholders were selected on the basis of their breadth of experience and knowledge, as well as their willingness and ability to articulate their opinions. Optimal sample size in research with the Delphi technique has not been established. Research has been published that was based on samples that vary from between 10 and 50 to much larger numbers (
63). Murphy and colleagues (
59) asserted that a larger sample is better, concluding that as the number of stakeholders increases, the reliability of "composite judgment" increases. However, these authors also stated that there is scant empirical evidence about the effect of the number of stakeholders on either the reliability or the validity of consensus processes.
The Delphi panel comprised seven stakeholder groups. It was our goal to have four participants from each of the stakeholder groups complete the Delphi process. The members of the stakeholder groups were selected from four levels: national, provincial, regional, and service. The expert group was selected at the national level and the payer group was selected from health ministry officials from the provincial government, because in Canada the provincial governments are responsible for funding and the delivery of health care. Two groups were selected at the regional level. The regional level consists of a single regional health authority, which is the health provider organization legally mandated to provide the continuum of health care services to the entire population in the region. In this case the regional health authority serves 1.2 million individuals. The two groups selected at the regional level were senior administrators and family physicians. Finally, three groups—families, consumers, and clinicians—were selected at the service level.
The seven stakeholder groups that formed the panel and the numbers of participants within each group are listed in
Table 2.
During the proposal-writing stage, the primary author contacted nationally recognized experts, government representatives from the Ministry of Health and Wellness (payer group), and administrative representatives from the provider organization to explain the project and details of participation (
Table 2). A list of family physicians that most frequently referred patients to the local early treatment program for psychosis was also identified. Potential family and consumer participants were identified by staff of the local early treatment program for psychosis.
The selection of the mental health care providers differed from that of the selection of the rest of the stakeholder groups. Rather than purposefully select four stakeholders, the research team invited all staff members of the local early treatment program for psychosis to participate in the Delphi technique and randomly sampled four of the seven participants' questionnaires for inclusion in our analysis.
The Delphi questionnaire was administered by the research coordinator in person to each member of the patient stakeholder group. All other stakeholders received a written questionnaire, either by e-mail or post. The stakeholders were asked to rate the importance of individual measures in the evaluation of quality of care in early psychosis programs. Each round of questionnaires included a qualitative component that offered the opportunity to provide additional feedback in the form of written comments. After round 2 and round 3, the degree of consensus achieved in the Delphi process was assessed by calculating the semi-interquartile range of the score assigned by the stakeholder for each measure (
54).
The semi-interquartile range is calculated as (75th percentile-25th percentile)/2.
The level of consensus was set before data were collected. Consensus was defined as being reached when measures attracted final scores with a semi-interquartile range of .50 (absolute). Measures with final scores with a semi-interquartile range of less than .50 were interpreted as having reached strong group consensus (
54).
Each round built on responses to the former round. Stakeholders were provided with a summary of the series of rounds. This summary included the feedback to each stakeholder: his or her own score on each item, the group's median ratings, and a synopsis of written comments. Stakeholders were then asked to reflect on the feedback and rerate each item in light of the new information.
In round 1, 25 stakeholders were asked to list five to ten performance measures that they believed to be important in the evaluation of the quality of care in early psychosis treatment services. The suggested performance measures were analyzed by using thematic content analysis with the Nud*ist (Non-numerical Unstructured Data Indexing Searching and Theorizing) computer software program (
64). This qualitative analysis was conducted only on this first round and not on subsequent rounds and resulted in the identification of 11 potential performance measures that were not identified by the literature review.
As shown in
Table 2, a total of 22 of the 25 original stakeholders participated in round 2. Three stakeholders withdrew, one from the payer group, one from the mental health administrative group, and one from the patient group. A questionnaire containing a comprehensive list of performance measures was distributed to participants. This list of 83 measures comprised the 73 items identified in the literature review plus ten additional items that were suggested in the first open-ended round of the Delphi process. Participants were asked to rate each of the measures on a 5-point Likert scale to determine the degree to which they thought the measure was essential.
Twenty of the 22 participants from the previous round participated in round 3 (
Table 2). One patient became ill and was hospitalized, and one stakeholder in the payer group withdrew from the study. Each performance measure was listed with the participant's own rating from the previous round, the median rating of the group, and the percentage of participants who responded to each rating on the Likert scale. Participants were asked to rerate each measure in light of this new information. In the event that their response was more than two points away from the group median, they were asked for elaborative comments.
Discussion
This is the first reported study to develop a set of performance measures specifically designed to evaluate early intervention services for psychosis. These measures are relevant for all mental health programs that provide services to individuals who experience a first episode of psychosis. They were not developed to evaluate only one specific model of early intervention services for psychosis; as a result, their validity does not depend on evidence from clinical trials or meta-analyses stating that one form of early psychosis is more effective than another (
66). Furthermore, the stakeholder consensus process established the face validity of these performance measures (
67). Publication of this set is timely because of the interest in innovative early intervention services for psychosis and the current lack of certainty about their superiority over treatment as usual (
4,
5,
6). As such the measures will be particularly relevant in the United States, where there has been less emphasis on the development of specific early-intervention services for psychosis.
The responses from the open-ended question in the first round of the Delphi technique were illuminating in that results indicated that different stakeholder groups tended to value different performance measures. For example, the experts emphasized the importance of access, perhaps reflecting their knowledge about the link between duration of untreated psychosis and outcome (
67); the payers emphasized readmission rates and costs; the patients emphasized management of side effects; and the families emphasized illness education.
Thirteen of the measures within the effectiveness domain were rated as essential or very important during the third round. This domain is particularly well developed and perhaps reflects the general trend in reporting outcome.
Given the diversity of the group, it is surprising that consensus was reached on a majority of measures in two rounds of the questionnaire (rounds 2 and 3). Future research could examine the reasons behind the opinion change. How participants behave between rounds and the reasons for their opinion change is an interesting psychological issue (
68). Other than in the first round, we did not conduct a thematic analysis on the qualitative responses, because it was beyond the scope of this study to analyze the effect of the qualitative comments on the subsequent decision making of stakeholder groups.
This study has a number of limitations. The measures rated were based on a literature search up to July 2002. The composition of stakeholder panels used in the Delphi technique is an important factor in judging the legitimacy of the findings, and some of the groups were drawn from only one region (
63). Attrition was a concern, such that the numbers in some of the stakeholders groups were not equivalent (
Table 2). Because of differences in group size in round 3, the results could be biased in favor of the mental health clinician providers and the family members.
At this stage of research the reproducibility of the results of this approach is not known. However, the Delphi technique used in this study has been clearly described and can be replicated by other investigators. Although the performance measures listed in
Table 4 represent consensus among the various stakeholders, some of the measures, such as knowledge and application of evidence-based practice and formal and continuing education of early psychosis treatment services staff—could be considered to be professional aspirations (
56).
Finally, although the significant benefits of the Delphi process are outlined above, the process itself has limitations. The major concern is that only limited feedback was included between rounds. Also, the process does not allow for face-to-face discussion, which is allowed by consensus development conferences (
69). Although other consensus techniques, such as the Nominal Group Technique RAND Appropriateness Method, have been used (
69), each technique has its own strengths and weaknesses.