Psychosomatics 39:311-317, August 1998
© 1998 The Academy of Psychosomatic Medine
Timing of Referral to a Consultation-Liaison Psychiatry Unit
Dennis Handrinos, M.B.B.S., M.P.M.,
Dean McKenzie, B.A., and
Graeme C. Smith, M.D.
Received June 30, 1997; revised December 3, 1997; accepted January 23, 1998. From the Consultation-Liaison Psychiatry Research Unit, Monash University, Department of Psychological Medicine, Clayton, Victoria, Australia. Address reprint requests to Dennis Handrinos, Dandenong Hospital, David Street, Dandenong, Victoria 3175, Australia.
|
ABSTRACT
|
This study examined ways of determining a useful measure of the timing of referral to a consultation-liaison psychiatry unit. Seven hundred and twelve consecutive inpatient referrals between February 1990 and January 1995 were studied prospectively. Usage of a simple mathematical ratio of the time elapsed between admission and referral and length of stay was shown to be independent of the patient's length of stay when this was greater than 4 days and thus is a more useful way of measuring timing than is the time elapsed between admission and referral. By using this measure, the authors found that a diagnosis of personality disorder predicted earlier referral, and depression cited as a reason for referral by the consultees predicted later referral. Independent of length of stay, the early referred patients had more time spent with them by the consultation-liaison unit.
Key Words: Consultation-Liaison Psychiatry • Referral • Timing
|
INTRODUCTION
|
As the cost of hospitalization comes under increasing scrutiny, ensuring a time-efficient hospital stay for patients becomes essential.1,2 If the psychiatric input is maximized, small reductions in hospital stay can have great multiplier effects on cost reductions and generally lead to better use of hospital resources.3 Determining guidelines for the optimal timing of referral to consultation-liaison (C-L) psychiatry is a logical goal. For some patients, for a variety of reasons, the need for referral to psychiatry may not be immediately apparent to referring physicians, and this delay may in turn adversely affect the length of stay of the patient. Indirect evidence for this view comes from studies1,3 that have shown that a patient's length of stay (LOS) can be reduced by having consultation-liaison psychiatry review all patients at admission. The conclusion to be drawn from this work is that to wait for the recognition of the need for referral, as is the practice in C-L psychiatry today, may unduly delay the discharge of the patient.
The timing of referral may have a variety of determinants. Some patients have the need for C-L psychiatry referral on admission; others may develop the need during their hospital stay. Some have the need recognized as it arises; in others there may be a significant delay.
There have been few published studies to date that focus on the issue of timing of referral. Ormont et al.4 investigated factors associated with the time of referral. In this study, time elapsed between admission and call for consultation (REFLAG) was used to indicate the timing of referral. The study found that the patients with schizophrenia were referred early and that the patients with AIDS (acquired immunodeficiency syndrome) were referred later in the admission. The study found a positive correlation between an earlier referral and earlier discharge from hospital.
Lyons and Hammer5 reported on 419 consults, for which they attempted to determine the relationship between the timing of referral and LOS. The group derived a "timing" variable according to the following formula:
timing variable=log (REFLAG)/log (LOS)
A linear regression equation for the relationship between this timing variable and LOS was then derived (below), which showed a positive association, explaining 12% of the variance between the timing variable and the eventual length of stay of the patient.
log LOS=0.447 * timing of referral+8.874
The same methodology was applied by Ackerman et al.6 The group studied 92 medical and surgical patients who were referred to the C-L service and who also met criteria for DSM-III depression. The researchers found a significant correlation between the timing variable and LOS and that the timing variable accounted for 21% of the variance in LOS. The main conclusions drawn from both papers reviewed by Ackerman et al. was that timing affects LOS and and that the patients referred early have shorter LOS.
To take this line of work further, we hypothesized that it may be possible to find a method for deriving a timing variable that was less dependent on length of stay. We further hypothesized that it would be possible to identify a meaningful and useful group of clinical indicators that could characterize the timing of the referral.
|
METHODS
|
The MICRO-CARES clinical database system7 was used to record details of all referrals made between February 1990 and January 1995 to the C-L Psychiatry Unit at Dandenong Hospital, a 300-bed general-teaching hospital affiliated with Monash University. At the time, the C-L Unit was staffed by a half-time consultant psychiatrist and a full-time psychiatry trainee. Time-process variables—LOS and date of referral, patient demographics, medical diagnosis using International Classification of Disease–9th Edition and psychiatric diagnoses in all five axes of DSM III-R, reasons for referral as given by the physician making the referral and also as conceptualized by the C-L unit, and recommendations made by the C-L Unit—were recorded for each consultation. Details of the use of the database system and reliability checks have been described by Smith et al.8
Statistical Analysis
To examine the relationship between referral lag and LOS, Pearson product moment correlation and linear regression were carried out by using SPSS/PC+.9 Such techniques look at the relationship between the two variables for the entire sample. However, it is possible that there are subgroups or clusters in the data with regard to relationships between referral lag and LOS. To investigate this, a cluster analysis with Bonferroni adjustment was performed by using the KnowledgeSEEKER (KS) computer program.10,11 KS searches for subgroups or clusters that are homogenous (have similar values) with relation to the dependent variable, in this case REFLAG. Clusters whose means are not significantly different from each other will be merged by the program. The analysis of variance (ANOVA) results are adjusted for the number of comparisons, using the Bonferroni technique. This technique has been applied to psychiatric research data by McKenzie et al. and Smith et al.12,13
|
RESULTS
|
The Study Group
There were 921 inpatient referrals in the 5-year period. Two hundred and nine patients referred following an episode of deliberate self-harm were excluded, as these patients are referred as a matter of hospital policy. It was felt that the policy would confound this study, which aimed to define and predict referral characteristics among physicians.
Seven hundred and twelve consults were available for analysis when those referred for self-harm were excluded. The mean ± SD LOS was 24.0 ± 24.3 days, compared with the mean ± SD LOS of all hospital patients of 7.73 ± 10.89 SD days. The mean ± SD REFLAG was 10.0 ± 12.8 days. The mean age of patients seen was 51 years; 53% were female and 70% of all consults were receiving some form of Social Security pension.
Part One: Finding a Relationship Between LOS and Timing of Referral
There was a strong correlation between LOS and REFLAG (r=0.6886, P<0.001). this relationship accounted for 47.4% (0.68862) of the variance in REFLAG. When the ranked values of LOS were analyzed, the correlation was slightly higher (r=0.6927, P<0.001), explaining 48.0% of the variance. Long LOS patients tended to have long referral lags, even though they had the opportunity to have been referred earlier in their hospital stay. Logarithmic transformation of the LOS, to account for the skewness, also remained highly correlated with REFLAG (r=0.6804, P<0.001) but accounted for slightly less variance (46.3%).
The data were then analyzed by using the KS program. As with the earlier correlation analyses, the KS program found that of all possible variables, REFLAG was most highly associated with the LOS variable. Consults were clustered into nine LOS groups (F=78.26, df=8,703, P<0.001) (Table 1).
The clustering into nine groups accounted for 47.1% of variance in REFLAG. As expected, the mean REFLAG in each group increases with the LOS. However, somewhat more important, the standard deviation also increases, indicating that not only is there a greater spread of REFLAGs in the higher LOS clusters, but that comparing low and high LOS clusters (or REFLAG groups) may be statistically invalid. The Levene test14 found the group variances to be highly different from each other (F(8,712)=117.61, df=8, P<0.001), which violates the ANOVA assumption of equality of variances across populations. However, the Brown-Forsythe15 modification of the ANOVA, which does not require variances to be equal, was statistically significant (F(4,38)=54.93, P<0.001).
The REFLAG variable is heavily influenced by the LOS, as evidenced by the high correlation and the results of the earlier KS analysis. Therefore, if the timing of a referral is to be the focus of study, the time elapsed between admission and referral may not be the best indicator of the timing of referral. It is merely an associated variable of a patient's LOS. A study looking at variations and determinants in referral lags, such as the study performed by Ormont et al.,4 could essentially be a study of variations and determinants of LOS in patients and would only partially address the issue of the timing of a referral.
It is for this reason that a simple ratio of the time elapsed between admission and referral and the LOS was proposed as a way of conceptualizing the timing of referral. Timing is mathematically expressed according to the following formula.
referral timing (RFtime)=REFLAG/LOS
The mean RFtime for the study cohort was 0.46, meaning that, on average, the patients were referred just before the half-way mark of the admission. Analyzing RFtime against LOS, the KS program split the study cohort into three LOS groups (F(2,709)=26.92, P<0.001) (see Table 2).
Even though the three groups differed on RFtime, according to standard one-way ANOVA (F(98,536)=6.23, P<0.001) and the nonparametric Kruskal-Wallis test16 ( 2=45.55, P<0.001), only 7.1% of the variance was accounted for, compared with 47% of the variance accounted for between the raw REFLAG and LOS.
A low correlation was found between RFtime and LOS. Although significant at P<0.001, only 2.70% (r=-0.1642) of the variance in LOS was accounted for by RFtime. When the 1–4-day LOS group was excluded (the ratio will vary greatly with small differences in timing in this short stay group), there was no significant correlation (r=-0.0911, P>0.01, variance=0.82%) between RFtime and LOS. Logarithmic transformation of RFtime and LOS also showed no significant correlation. This is despite RFtime being partially derived from the LOS variable. The use of RFtime appears to equalize the LOS group variance. RFtime and LOS are not significantly related in patients, especially when LOS was greater than 4 days.
Part Two: Finding a Relationship Between Timing of Referral and Clinical Variables
As RFtime was shown to account for only a small percentage of the variance of LOS, further analyses were aimed at examining whether the timing of referral (RFtime) would correlate with other measured clinical factors and hence be a useful measure or predictor. On the basis of the aforementioned findings, further analysis was performed after excluding the 1–4-day LOS group; 621 consults were available for this analysis. Analyzing each variable separately, RFtime had significant associations with two variables: 1) the total number of visits performed by the C-L unit and 2) when "depression" was given as a reason for referral by the physicians (consultees) (see Table 3).
The higher the RFtime ratio, the fewer visits were performed. When depression was given as a reason by the consultee, the patient was referred later in the admission. These findings contrast with many more associations found between the referral lag and consultation variables (see Table 4), indicating the importance of defining the method of measuring timing before attempting to deduce its determinants.
Of note, the only association found from using either timing variable was in consults when depression was given as a reason for referral by the consultee.
KS analysis was then performed to provide a multivariate profile of the early or late referral according to the RFtime value. As mentioned earlier, KS found three LOS groups when analyzing RFtime against LOS (see Table 2). The 8–156-day group (N=544) was used for further analysis. The 5–7-day group could also have been used, but it contained an insufficient number of consults to perform this form of analysis adequately. The KS program allows for a variety of models to be constructed with differing groups of variables, each model maintaining grouping significant at P<0.01. Consistent with the aforementioned bivariate analysis, the total number of visits performed by the C-L unit was by far the best initial discriminator of early and late referral.
An alternate model was constructed by using consultee-cited "depression" as the main discriminating variable. The program discriminated between two groups, such that when depression was given as a reason for referral, the mean RFtime was 0.46 (SD=0.26, N=236), and if not jgiven, the mean RFtime was 0.4 (SD=0.28, N=308). This was significant at P<0.01. The model also revealed that in both groups, the earlier the unit referred the patient, the more time was spent or the more visits were performed. In this model, a cluster of a positive personality disorder diagnosis in the absence of depression as a reason for consultation yielded the lowest mean RFtime of 0.28. Other than time spent and associated variables, and personality disorder diagnosis or psychotic prior history as a reason for referral, there were no groupings or clustering of variables that could discriminate between late or early referral. Of note, variables that were found to have no bearing on RFtime included the age, gender, and employment status of the patient; the unit making the referral; the urgency of the referral; psychiatric and medical diagnosis of the patient; and discharge location of the patient.
|
DISCUSSION
|
This study has tried to derive a valid timing variable and then to describe significant factors associated with variation in the timing of referral. As the time elapsed between admission and referral and the LOS are closely correlated, a study looking at timing has to deal with the LOS variable in a way that does not make it the center of investigation. Late referral would also mean long stay; early referral would also mean short stay. Conceptualizing timing of referral in a way that is independent of the LOS is considered an improvement to the raw REFLAG timing variable and is the first major finding in this study. We found that the simple ratio of RFtime did not significantly covary with LOS, particularly in the patients with a stay greater than 4 days. Such loss of information in representing such variables by a simple ratio may have caused the low correlation between LOS and Rftime. Therefore, in this study, for the LOS group as defined, timing of referral can be usefully conceptualized as an event occurring in relation to the total proportion of the hospital stay rather than just the time elapsed.
The second major finding of this study is that if patients were referred early in proportion to the total time they spent in hospital, the C-L unit assigned more time to the consult and visited the patient more often. As RFtime variable was independent of the LOS, so the finding that more time was spent on these patients was independent of the total LOS. It appears that patients referred earlier in their admission required and received more overall involvement by the C-L unit in their care. This was independent of the total days spent in hospital, which would merely reflect more opportunity for the C-L unit to devote time and resources to the patient.
We found that patients referred because of consultee-cited depression were referred late. Of note, this association was not found with the diagnosis made by the consultant of a mood disorder. The finding is more indicative of the language used by consultees when referring a patient rather than the final diagnosis. Whatever it was that the consultees were detecting and referring as depression, be they depressed patients or those with delirium or other psychiatric disorders, these consults were being referred late. Whether this need for referral was present on hospital entry or whether it developed in the hospital stay could not be studied by using the present study design. Similarly, the severity of physical illness and its effect of LOS and timing of referral could not be measured in detail in this study.
A drawback of defining timing with a ratio of REFLAG to LOS is that both short and long LOS patients are treated as if they are equal. For example, for a RFtime of 0.3, a referral on Day 3 with a LOS of 9 days is treated the same as a referral on Day 30 with a LOS of 90 days. Assuming that these admissions have similar characteristics and comparing them as a group with admissions with RFtimes of, say, 0.9 may be invalid. A similar drawback is that the RFtime method of timing does not readily allow for measurement of any influence the referral to psychiatry may have had on expediting an earlier discharge for that particular patient. Necessarily, low RFtime admissions would spend a greater proportion of time in hospital following referral, and high RFtime admissions would be discharged soon after referral. To overcome these difficulties, a future study could confine the study cohort to a predetermined LOS, say 7–14 days, and then investigate variations within RFtime in this group. Similarly a study could examine differences between two LOS groups with predetermined RFtimes.
More work is required to replicate this proposed method of conceptualizing the timing of referral, particularly on differing patient cohorts in other hospitals. Further timing studies investigating the specific discrepancy between the origin of the need for referral and the eventual recognition for referral seems the next logical step in this line of research. Reduction of unnecessary delays in the provision of patient care is where ethical efficiency gains may be found.
|
ACKNOWLEDGMENTS
|
The authors thank the psychiatry trainees Drs. Jack Kirzenblat, David Recht, Hugh Lowy, and David Sholl, who diligently collected the data; Ms. Penelope Liapis for manuscript preparation; and the Buckland Foundation and Upjohn Australia for financial assistance.
|
REFERENCES
|
-
Levitan SJ, Kornfeld DS: Clinical and cost benefits of liaison psychiatry. Am J Psychiatry 1981; 138:790–793[Abstract/Free Full Text]
-
Fulop GF, Strain JJ, Vita J, et al: Impact of psychiatric comorbidity on length of hospital stay for medical/surgical patients: a preliminary report. Am J Psychiatry 1987; 144:878–882[Abstract/Free Full Text]
-
Strain JJ, Lyons JS, Hammer JS, et al: Cost offset from a psychiatric consultation-liaison intervention with elderly hip fracture patients. Am J Psychiatry 1991; 148:1044–1049[Abstract/Free Full Text]
-
Ormont MA, Weisman HW, Heller SS, et al: The timing of psychiatric consultation requests: utilization, liaison, and diagnostic considerations. Psychosomatics 1997; 38:38–44 [Abstract/Free Full Text]
-
Lyons JS, Hammer JS: The timing of psychiatric consultation in the general hospital and length of hospital stay. Gen Hosp Psychiatry 1986; 8:159–162[Medline]
-
Ackerman AD, Lyons JS, Hammer JS, et al: The impact of coexisting depression and timing of psychiatric consultation on medical patients 'length of stay. Hosp Community Psychiatry 1988; 39:173–176[Abstract/Free Full Text]
-
Hammer JS, Strain JJ, Lyerly M: An optical scan/statistical package for clinical data management in C-L psychiatry. Gen Hosp Psychiatry 1993; 15: 95–101
-
Smith GC, Clarke DM, Hermann HE: Establishing a consultation-liaison psychiatry clinical database in an Australian general hospital. Gen Hosp Psychiatry 1993; 15:243–253[Medline]
-
Norusis MJ, SPSS Inc.: SPSS/PC+4.0. Base manual for the IBM PC/XT/AT and PS/2. Chicago, IL, SPSS Inc., 1990
-
Biggs D, De Ville B, Suen E: A method of choosing multi-way partitions for classification and decision trees. Journal of Applied Statistics 1991; 18:49–62
-
Angoss Software International: KnowledgeSEEKER User's Guide. Toronto, Ontario, Canada, Angoss Software International, 1993
-
McKenzie DP, McGorry PD, Wallace CS, et al: Constructing a minimal diagnostic decision tree. Methods Inf Med 1993; 32:161–166[Medline]
-
Smith GC, Clarke DM, Handrinos D, et al: Consultation-liaison psychiatrists' management of depression. Psychosomatics 1998; 39:244–252[Abstract/Free Full Text]
-
Levene H: Robust tests for equality of variance, in Contributions to Probability and Statistics, edited by Olkin I. Palo Alto, CA, Stanford University Press, 1960, pp. 278–292
-
Brown MB, Forsythe AB: Robust tests for the equality of variances. Journal of the American Statistical Association 1994; 69:364–367
-
Siegel S, Castellan NJ: Nonparametric Statistics for the Behavioural Sciences, 2nd Edition. New York, McGraw-Hill, 1988
This article has been cited by other articles:
|
|
|
|
 
Y. Kishi, W. H. Meller, M. Kato, S. Thurber, S. E. Swigart, T. Okuyama, K. Mikami, R. G. Kathol, T. Hosaka, and T. Aoki
A Comparison of Psychiatric Consultation Liaison Services Between Hospitals in the United States and Japan
Psychosomatics,
December 1, 2007;
48(6):
517 - 522.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
Y. Kishi, W. H. Meller, R. G. Kathol, and S. E. Swigart
Factors Affecting the Relationship Between the Timing of Psychiatric Consultation and General Hospital Length of Stay
Psychosomatics,
December 1, 2004;
45(6):
470 - 476.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
P. de Jonge, F. J. Huyse, G. M.-F. Ruinemans, F. C. Stiefel, J. S. Lyons, and J. P.J. Slaets
Timing of Psychiatric Consultations: The Impact of Social Vulnerability and Level of Psychiatric Dysfunction
Psychosomatics,
December 1, 2000;
41(6):
505 - 511.
[Abstract]
[Full Text]
|
|
|
Get information about faster international access.
Privacy Policy
Copyright © 1998
Academy of Psychosomatic Medicine.
All rights reserved.
Home
| Search
| Current Issue
| Past Issues
| Subscribe
| All APPI Journals
| Help
| Contact Us
|
TOP
ABSTRACT
INTRODUCTION
