The abstracts from steps (d) and (e) were screened with the purpose of identifying literature describing the frontier of knowledge about the properties and modifications/changes of GAF. The frontier of knowledge is the boundary or limit of current knowledge. When this screening started, the researcher was experienced from reading literature from steps (a) to (c). Abstracts were evaluated for inclusion by looking for information on the following issues in relation to GAF: scaling, nature of anchor points, scoring of symptoms and functioning, scoring within 10-point intervals, psychometrics (studies with information on validity and reliability), history of GAF, modifications/changes made, and a more multidimensional GAF. When the screening of abstracts was finished, selected publications were read in their entirety, but it became clear that most of the relevant literature had already been identified by steps (a) to (c).

A continuous scale has no steps and does not force the respondent to answer in specific categories [52]. In GAF, a continuous scale (finely graded with 100 points) has been preferred to a discrete scale. With good reliability, sensitivity using continuous scales can be good for detecting change and differences. Statistical testing can show statistically significant differences for samples with small differences in the severity of illness. Continuous scales may also be applied to defining threshold values for assigning diagnoses. It is plausible that symptoms and functioning are more continuous in nature than mental illness itself. Error of measurement for such a finely graded scale may also mask a possible discontinuity of mental disorders. In GAF, the anchor points are ranked, but it is open to question whether the anchor points (with key words and examples) really constitute a natural continuum.

A VAS (visual analogue scale) is a line with anchor points at each end to indicate the extremes. The scorer marks a point on the scale indicating the severity of the phenomenon. The scored value is the distance from the point to the scale's lower end. The VAS has been used successfully in psychiatry, but there is no conclusive evidence that it is better than categorical scales and it takes more work to analyse [46,51,53,54,61,62]. When a VAS is equipped with descriptive anchor points along the line, it becomes more similar to a scale that could work as a visual scale for GAF. Technologically, it is possible to computerise scoring on a VAS by setting a mark on the screen's digital line, so the computer calculates the distance from the lower end of the line.

Raw data from scaling and measurement often undergo statistical analysis. For such analysis, it is relevant to distinguish between four types of scales: nominal, ordinal, interval and ratio scales. Both nominal and ordinal scales are well known in psychiatry and GAF is an example of an ordinal scale. This has consequences for further treatment of data. We cannot say, for example, that a 5-point change in GAF from 38 to 43 means the same change in severity as that from 68 to 73. Mean GAF at the start of treatment minus mean GAF at the finish, for sample A, cannot be said to be larger than the same change for sample B, in spite of sample A clearly having a larger numerical difference than sample B [22]. Similarly, it is not entirely correct to add individual scores and divide by the number of individual scores to obtain the mean value. For psychiatry, it is difficult to develop a mental health scale that reaches the level of a real interval or ratio scale, but it is quite common to see GAF data treated as something more than ordinal data. In some research projects, collected raw data for GAF are merged into a limited number of categories [15,63]. A simple version of this is to dichotomise the level of functioning into 'superior to fair' and 'poor to grossly impaired' [64]. Some authors have merged their raw data into more categories (from three to seven [15,63,65-67]). It would be expected that such categorisation of a raw data set is important for conclusions drawn when the data are treated statistically. For a single scale GAF 'whichever is the worse' of an individual's symptom and functioning values is the GAF score [68]. Also, when scoring is performed on two separate scales (GAF-S and GAF-F scales), sometimes only one score is recorded. In principle, this could be the lower, average or higher of the two scores. As GAF-S and GAF-F score different dimensions, giving just one figure is open to criticism and also means loss of information.

The 10 anchor points (with key words and examples of symptoms and functioning items) give a general idea on what to stress in scoring GAF. The use of examples is important and is likely to improve assessment [73]. In Hall's 'modified GAF' a greater number of criteria for scoring are found [28]. Items used in different symptom and functioning scoring systems are different; in further work with GAF, ideas for the best subset of items can be drawn from the literature on symptom and functioning scoring [2,22,53,74,75].

The current symptom anchor points were generally assigned in earlier stages of development that led to the present GAF, but much symptom research has been performed since then. Symptom checklists can include questions about behavioural and somatic symptoms, and positive and negative feelings of well-being [22,76]. Asking about both positive feelings of well-being and somatic symptoms makes the checklist more objective; sensitivity and specificity can be good, and the intent of the measurement is concealed [22]. As patients can have more than one symptom, with different types and degrees of development, assessments of illness severity based on such symptom clusters seems logical. Many symptoms in psychiatry have two aspects: form (for example, auditory hallucination) and content (for example, the person is told to do something) [77]. In symptom-scoring systems, symptom content has been largely ignored, but perhaps it should not be [73].

A large number of indices of functioning have been constructed [17,22,74,78]. Functional status can be defined as the degree to which an individual is able to perform socially allocated roles free of mentally (or physically) related limitations [74]. A measure of functioning requires decisions about: which type of functioning should be scored (for appraisal of overall functioning, several types of functioning should be scored, for example difficulties with participation in working life, daily activities, and social relationships); how to grade each type of functioning; and whether an aggregate measure can be made (that is, the total score expressed with one figure).

In psychiatry, there is a preoccupation with mental illness, but less interest in positive mental health [70,79]. Positive and negative feelings are not simply opposite ends of a single-dimension scale [22]. It could be discussed whether the scoring of GAF should include factors such as life satisfaction, positive quality of life, psychological well-being, and even physical fitness [70,71,74]. Inclusion of questions about 'positive mental health' may be important for prediction of the ability to improve after an episode of mental illness.

The present GAF has limited value for assessing prognosis [63], and other systems predict prognosis better [25,36,53]. Prognosis is definable as a part of the severity of illness. A patient who is severely ill with a good prognosis can then be scored more highly than a patient who is less severely ill with a poor prognosis. Prognosis can be related to the patient's resources and not just the patient's problems and is more dependent on diagnosis and symptoms than impairment ratings: the highest level of functioning for a time period is more important for prognosis than the lowest, and substance abuse plays a role [15,70,71,74].

In the DSM-IV-TR, there is an overlap between criteria for diagnoses and criteria for GAF scoring. A relationship with diagnoses can be expected for GAF [15,26,32,34,63,80,81], but DSM is a multiaxial system [32] where each axis is intended to add information. In their work with GAS, Endicott et al. [18] wanted to remove all diagnostic criteria. A different strategy would be to develop different criterion sets for different diagnoses (for example, for dementia and depression). The use of diagnosis-specific symptoms and functioning criteria for GAF scoring could improve the generic properties of GAF.

The anchor points are generally inclusion criteria for scoring in 10-point intervals. Little work has been performed to identify exclusion criteria for scoring in each interval. An example would be identification of symptoms (or grading of symptoms) that exclude scoring in the GAF-S interval 51-60 and make the interval 41-50 preferable. Proposing that the anchor points of neighbouring 10-point intervals are exclusion criteria may be too simple an answer.

The GAF scale identifies the lowest and highest levels for a hierarchy of mental illness. The choice of anchor points at the endpoints is decisive for the variation in possibilities of a phenomenon, as endpoints can influence which score is given [62]. In scoring of morbidity, perfect health often marks one extreme. In GAF-S, the other extreme is persistent danger of severely hurting themselves or others, and in GAF-F it is persistent inability to maintain minimal personal hygiene. In a disease-staging system, death was chosen as the lower endpoint for a number of psychiatric conditions [55]. However, not all health states can be placed upon a continuum bounded by the anchor points 'perfect health' and 'death' [62]. Patients themselves can consider some conditions worse than death [52,62]. In the Kennedy Axis V's subscale for psychological impairment, criteria have been added to the GAF criteria, such as 'totally insensitive to the feelings and need of others' (the lowest interval) [83]. The first step in work with a scaling instrument should be to define its endpoints.

The 100 scoring possibilities in GAF and the low detail of verbal instructions are in conflict with each other. Equipping GAF with a higher number of anchor points could be considered [10]. In general, the middle range is frequently used in psychiatry, and more elaborate verbal instructions for the middle range could be considered [82]. For newly admitted inpatients, higher scorings are rarely used, which gives relevance to having more anchor point for the lower range [18]. In community studies, the upper part of the scale is most relevant, and so the question of having more anchor points for the upper range also comes up. When scoring of GAF is computerised, links can be visible on the screen and clicking on these links gives more detailed information (for example, for scoring newly admitted inpatients and for community studies).

Endicott et al. [18] and the manual for DSM-IV-TR give instructions for scoring within 10-point intervals, but instructions are limited. In practice, clinicians tend to score around the decile, or mid-decile, divisions of the scale [16]. When information for a more accurate score is lacking, intermediate scores in the deciles are chosen [21,51].

An alterative procedure for scoring within 10-point intervals is found in the 'modified GAF' [28]. The number of criteria met is used, for example for the interval 41-50: when one criterion is met the score should be 48-50 and when two criteria are met the score should be 44-47.

Reliability and validity studies for both GAF-S and GAF-F scales exist, but there are relatively few [2,8-10,15,26,30]. In psychiatry, symptoms and functioning are often closely related [15,17,26,63], but have been proposed to deviate frequently enough to recommend measuring both in outcome studies [17,35]. Functioning can improve without a corresponding symptom improvement and vice versa [35]. GAF-S and GAF-F can be correlated with r = 0.61 [10]. When GAF-S scores share more variation with other measures of symptoms and GAF-F scores share more variation with other measures of functioning [10], this suggests that GAF-S and GAF-F represent different aspects of a patient's condition. Few studies have focused on concurrent validity of GAF-S and GAF-F separately, but the association between GAF-F and other types of functioning may be low [10,15,30,63]. In general, we have little empirical knowledge about the advantage of separate scores for symptoms and functioning, for example, for assessment of treatment need and measurement of outcome [10]. The clinical significance, when GAF-S and GAF-F are clearly different, has also been little explored.

In the latest version of the DSM (DSM-IV-TR), two extra scales were provided for further study: the Global Assessment of Relational Functioning Scale (GARF) and the Social and Occupational Functioning Assessment Scale (SOFAS). The Mental Illness Research, Education & Clinical Center (MIRECC) GAF has three scales: for symptom severity, occupational functioning, and social functioning [8]. In the Kennedy Axis V, the seven subscales provide a broad profile of the patient [83]. GARF, SOFAS [5,26,29,86], MIRECC GAF [8], and Kennedy Axis V [83] all make more information available to the clinician. If the number of scales is increased, there may be a longer learning time for the scoring method, scoring becomes more time consuming and less easy to use, with analysis of the results becoming more complex (for example for outcome). International diffusion of these scales has been modest.

Researching the frontier of current knowledge and gaps in knowledge is a well known starting point for any study. Existing international research on GAF is characterised by researchers paying attention to some aspects (for example reliability), but there is less evidence of well thought out overall research programmes where different properties are systematically changed and tested in order to obtain an optimal system. In such research, independent variables can be different changes in properties, and dependent variables measures of reliability, validity and sensitivity. As GAF is intended to be a generic system, the work could be performed for different diagnostic groups. Although Hall [28] showed that changes in properties can improve GAF, it is not a matter of course that research where properties are changed results in an improved system. The simplicity of GAF is an advantage and a future GAF could become more complex. The potential gains with an improved GAF should be balanced against the consequence of a more time-consuming scoring for each patient (that is, a reduction in total capacity for the mental health service). Comparison between a new GAF and the current GAF will not necessarily show scores that are directly comparable [96]. This may be a problem for comparison of results from different studies, meta-analyses and use of historical data.