Objective - To determine
which characteristics were the best predictors of high rates of prescribing of glyceryl
trinitrate buccal tablets.
Design - Practice and patient characteristics
from 197 practices were examined, and a multiple regression analysis was performed to
examine which variables were important in predicting this prescribing.
Setting - Former family health services authority (197 practices).
Main outcome measure - Volume of prescribing of glyceryl trinitrate buccal
tablets.
Results - Four variables contributed significantly to a multiple
regression model: the catchment area of the secondary care establishment; the number of
partners in a practice; the level of practice deprivation; and whether the practice served
an urban or a rural area. The model suggests that the most important variable was the
catchment area of the secondary care establishment in which the practice was located.
Conclusion -Although only the prescribing of short acting glyceryl
trinitrate buccal tablets was studied, an impact of this size on primary care prescribing
may have extensive implications for all drug expenditure in primary care.
Dr Watson, in the lounge with the British National Formulary
Reclined on
the chaise longue, Dr Watson ran an interested eye through the November issue of the
Association of Cardiologists and Myocardial Experts. He was sipping his second cup of
Earl Grey when it caught his eye-the prescribing of glyceryl trinitrate preparations in
Buccalshire Family Health Services Authority differed greatly from the rest of the West
Midlands region (fig 1)!
Fig 1 - Prescribed defined daily doses of glyceryl
trinitrate formulations in family health services authorities of West Midlands region in
third quarter of 1994-5. Source: prescribing analysis and cost (PACT) data
There was a far greater use of glyceryl trinitrate buccal tablets
in Buccalshire: 316 defined daily doses (the average adult maintenance dose of a chemical
substance, as defined by the World Health Organisation(1)) per 1000 prescribing units
(practice list size adjusted to compensate for patient age(2)) compared with a mean of
34 defined daily doses per 1000 prescribing units for the rest of the
region!
He read on, fascinated. This difference did not seem
to be explained by higher morbidity rates (fig 2).
Fig 2 - Admissions into hospital per 100 000 prescribing
units for angina in family health services authorities of West Midlands region in 1993-4.
Source: Korner database
What was going on? He knew that
glyceryl trinitrate was an effective drug for providing both rapid symptomatic relief of and
prophylaxis against angina and also that there was no evidence to suggest that the
considerably more expensive modified release buccal tablets offer any clinical advantage
over the conventional short acting preparations-that is, sublingual tablets and aerosol
sprays. Although he knew that the buccal formulation did have a place in prophylaxis against
exercise induced angina,(3) he fondly recalled an analysis of glyceryl trinitrate
delivery systems which stated that it mattered little which glyceryl trinitrate preparation
was used so long as an adequate dose was given.(4) Slowly and deliberately he placed
the journal on the cocktail cabinet, deep in thought.
There was no alternative - he would
have to enlist the help of a statistician.
Miss Scarlett,
in the study with spss (for Windows version 6)
"I suppose that you want me to collect
prescribing analysis and cost (PACT) data on glyceryl trinitratebuccal tablets for all
Buccalshire practices (for the quarter ending November 1994), calculate the defined daily
doses to measure the volume of prescribing, and then analyse it all?" she said in one
breath.
"Yes," Dr Watson replied, "and remember to examine independent variables, such
as practice and patient characteristics, to control for variations in demand."
"Don't
worry," she trilled, "I'll include practice characteristics that have previously been
identified as potential influences on prescribing, and a measure of morbidity."
Playing
absent-mindedly with her pencil, Miss Scarlett pondered the problem of controlling for age
and sex. The age-sex distribution of the patient population is an important factor in
prescribing analysis, and various weightings such as the prescribing unit,(2) the age,
sex, and temporary resident prescribing unit (ASTRO-PU),(5) and the specific
therapeutic group age-sex prescribing unit (STAR-PU)(6) have been created to account
for these variations. However, she mused, all three weightings are based on, at the very
least, an entire therapeutic group. Therefore, use of the ASTRO-PU or STAR-PU at subsection
level seemed inappropriate. The division of age into under 65 and 65 years and over in a
prescribing unit does not accurately mimic the age-sex distribution of patients with angina:
the prevalence of angina rises dramatically from the age of 45 to 85 and then falls, with
men showing a sharper gradient across all age groups.(7) She decided to use practice
list size as the common denominator.
Fundholding has been shown to alter general
practitioners' prescribing habits,(8) and the number of partners in a practice has
been shown to determine the number of drugs prescribed.(9) Other factors she
considered important in affecting prescribing were the training status of the
practice(10) and whether or not it was a dispensing practice.(11)
Hospital
led prescribing also affects prescribing in primary care.(12) A survey of 240 general
practitioners in the West Midlands region showed that 76% stated that consultants' influence
was either "sometimes great" or "great."(13)
How could we measure this?
she wondered.
Colonel Mustard, in the library with the geographical information
system (GIS)
"Easily, my dear," boomed Colonel Mustard while sipping a large port. "We
calculate the Euclidean (or "as the crow flies") distance of each practice from the
nearest main acute secondary care unit, and so give each of the four main acute hospitals a
catchment area. I accept that this is only a proxy measure and that patients need not
necessarily be admitted to the nearest hospital, but the widespread urban geography of
Buccalshire Family Health Services Authority would make this a logical assumption."
"We
must also consider whether a practice is based in an urban or rural location," he
continued. "The number of general practice consultations is related to whether the patient
lives in a rural or urban area."(7)
"And deprivation of the patient population
has been cited as a potential determinant in variation in prescribing,"(14) added
Miss Scarlett. "How can we account for that?"
"We can measure patient deprivation
using the authority's patient registers and census data from the Office of Population
Censuses and Surveys to calculate Townsend scores(15) for each practice, although this
method has limitations."(16t)
"What is the Townsend score?" she asked.
"It's
a material deprivation index based on four census variables," he explained: "unemployment,
car ownership, the level of overcrowding, and the number of owner occupied properties. The
index is an unweighted combination of the four scores once the skewness of the unemployment
and overcrowding is reduced by means of the natural log transformation. This score performs
well in explaining variations in health."(17)
The clock on the mantelpiece ticked
ominously.
"The morbidity of the patient population, as measured by the number of
admissions to hospital, must also be included, my dear." He drained his glass, "Can I
interest you in a large one?"
Colonel Mustard and Miss Scarlett (the morning after),
in the study with a large bottle of analgesics
Sat side by side at the mahogany desk, the
two worked in silence. There were 197 practices in Buccalshire during September to November
1994. Of these, 63 practices did not fall into the four catchment areas and 10 practices
were atypical (university health centres and practices dealing only with care of patients
with terminal disease). Exclusion of these two subsets of general practitioners resulted in
the data for 129 practices being considered by Miss Scarlett.
Univariate Analysis
Miss Scarlett examined the data carefully for clues. Where evidence suggested that the
variable showed a trend, the two groups were compared with the Mann-Whitney U test-for
example, comparison of the average level of prescribing between fundholding and
non-fundholding doctors. Mann-Whitney U tests were also used to compare general
practitioners who prescribed glyceryl trinitrate buccal tablets with those who did not-for
example, to see whether there was a difference in either the median number of partners or in
the median practice list size between these two groups. Univariate analysis of catchment
area was by Kruskal-Wallis one way analysis of variance.
Multivariate Analysis
"However, some variables may only seem to be significant
because they are correlated with other variables that are significant," she said.
Colonel Mustard raised an eyebrow, "Really?"
"Yes," she nodded, "to identify
variables that are independently related to the volume of prescribing of glyceryl trinitrate
buccal tablets we must examine simultaneously all the variables in a stepwise multiple
linear regression analysis."
She nonchalantly selected the variables using two
probability criteria: the probability of the F statistic to enter the model was set
at a maximum of 0.05 and the probability of the F statistic to exit at a maximum of
0.10.
| Table 1 - Results of univariate analysis of data on 129 general practices in Buccalshire Family Health Services Authority |
| Factor |
Average (spread) |
Range |
Test statistic |
P value |
Correlation coefficient (P value) |
| Median No (lower and upper quartile) of partners |
2 (1, 5) |
1 to 8 |
U = 880.5 |
0.41 |
rs) = -0.47 (<0.01) |
| Mean (SD) Townsend score* |
0.27 (0.88) |
-1.82 to 2.73 |
NA |
NA |
r = 0.26 (0.003) |
| Mean No (SD) of admissions per 1000 practice list size* |
1.44 (0.91) |
0 to 5.63 |
NA |
NA |
r = 0.26 (0.003) |
| Median (lower and upper quartile) practice list size |
4451 (2337, 7975) |
276 to 18 141 |
U = 717.0 |
0.06 |
rs) = -0.42 (<0.01) |
| NA = not applicable. *Townsend score had single step function and so was dichotomised into high and low. Admissions seemed to be uniformly distributed. Townsend score and admissions variables both seemed to be normally distributed, and so means and standard deviations are quoted. |
Colonel Mustard and Miss Scarlett (some time later), in the ballroom with the
results
"I've put the results of the univariate numerical tests in table 1 and for
univariate categorical tests in table 2," said Miss Scarlett.
"Nice candelabras," observed Colonel Mustard.
"Does X mark
the spot?" (see table 2) he asked.
| Table 2 - Results of univariate analysis of data on 129 general practices in Buccalshire Family Health Services Authority |
| Factor |
Median defined daily dose per 1000 practice list size (lower quartile to upper quartile) |
Test statistic |
P value |
| Fundholding status: |
|
U = 1770.5 |
0.15 |
| Fundholding (n = 62) |
63 (4 to 229) |
|
|
| Non-fundholding |
125 (22 to 305) |
|
|
| Training status: |
|
U = 908.5 |
<0.01 |
| Training (n = 30) |
9 (2 to 88) |
|
|
| Non-training |
131 (17 to 305) |
|
|
| Dispensing status: |
|
U = 396.5 |
0.03 |
| Dispensing (n = 11) |
7 (0 to 136) |
|
|
| Non-dispensing |
106 (8 to 284) |
|
|
| Urbanity: |
|
U = 702.5 |
<0.01 |
| Urban (n = 107) |
120 (9 to 305) |
|
|
| Rural |
19 (0 to 113) |
|
|
| Catchment area: |
|
|
<0.001 |
| X (n = 87) |
198 (103 to 358) |
chi 2 = 64.17* |
|
| Y (n = 12) |
7 (1 to 36) |
|
|
| Z (n = 19) |
4 (0 to 7) |
|
|
| W (n = 11) |
3 (0 to 9) |
|
|
| Townsend score: |
|
|
|
| "High" (n = 64) |
200 (101 to 364) |
U = 1023.0 |
<0.01 |
| "Low" |
13 (2 to 130) |
|
|
| *Kruskal-Wallis one way analysis of variance. |
"Univariately, yes," she replied.
Initial
multivariate examination showed that there was indeed a spanner in the works, and three
practices were excluded as outliers. All three were single handed, non-dispensing,
non-training urban practices from catchment area X. One practice was fundholding and also
came from an area of "low" deprivation. The practices had comparatively small list sizes
(1746, 648, and 276) and admission rates of 0.57, 1.54, and 3.62 per 1000 practice list size
respectively. For the quarter studied they prescribed 582.47, 1915.12 and 2626.81 defined
daily doses of glyceryl trinitrate buccal tablets per 1000 practice list size.
The
residuals were examined for violation of the normality assumption. There was no evidence (P
= 0.32, Kolmogorov-Smirnov Z = 0.96) to suggest that the residuals did not follow an
approximate normal distribution, and all of the standardised residuals were within three
standard deviations.
"So, was it fundholders in rural locations with low
deprivation?" asked Colonel Mustard.
"No," said Miss Scarlett firmly. "See for
yourself. The resulting multivariate model is in table 3."
| Table 3 - Results of multivariate analysis of prescribing of glyceryl trinitrate buccal tablets in 126 general practices |
| Variable |
Regression coefficient |
95% Confidence interval |
Cumulative % of variance explained |
Feature associated with high prescribing rate |
| Constant |
345 |
212.43 to 479.04 |
- |
- |
| Catchment area |
-67.79 |
-91.53 to -44.06 |
28.4 |
Catchment area X |
| No of partners |
-29.93 |
-42.10 to -17.77 |
43.0 |
Fewer partners |
| "Townsend score" |
61.57 |
11.11 to 112.03 |
45.7 |
"High" deprivation score |
| Urbanity |
-64.76 |
-125.88 to -3.63 |
47.2 |
Urban practices |
"What about the corresponding hospital prescribing rates for that quarter?" he asked.
"Got that," she replied, handing him a sheet of neatly typed A4 (table
4)
| Table 4 - Drug company sales data for glyceryl trinitrate buccal tablets |
|
Total defined daily doses |
Defined daily dose per 1000 practice list size* |
| Hospital:
|
| X |
17 520 |
42.56 |
| Y |
1 800 |
17.17 |
| Z |
0 |
0 |
| W |
80 |
1.04 |
| *Of catchment area. Note: Figures do not necessarily reflect general purchasing trends of the hospitals because of bulk buying. |
Miss Scarlett, Colonel Mustard, and Dr Watson, in the
dining room with dinner
Colonel Mustard was stood at the head of the table, ladle in
hand.
"Anyone for broth?" he asked. "It's a real pea souper!"
"Thank you," said
Dr Watson, turning to Miss Scarlett. "Well, my dear?"
"The final model comprised four
variables," slurped Miss Scarlett: "catchment area, urbanity, the number of partners in a
practice, and the practice 'Townsend score.' This model explained over 47% of the variation
in the prescription of glyceryl trinitrate buccal tablets. The most influential variable
(chosen in the first step of the multivariate analysis) was catchment area: this variable
alone accounted for over 28% of the variation. It was admittedly used as a proxy measure for
the influence of secondary care establishments on the prescribing practice of general
practices.
"Urban practices in areas of 'high' deprivation were also likely to have
higher prescribing rates for glyceryl trinitrate buccal tablets, especially when there were
only a few partners in the practice. Morbidity, in terms of admission data, was not included
in the final model. This may be because the dichotomised deprivation score could be thought
of as a proxy measure of morbidity due to its association with variations in
health."(15) She dabbed delicately with her knapkin.
"Clearly, the model only
includes the variables considered-these may mask other underlying causes-and other
variables that we did not consider could be important factors in the high prescribing of
glyceryl trinitrate buccal tablets. Age is one such factor, and, although there was no
suitable method for weighting the age distribution for patients with angina, it is accepted
that this variable may also be an important predictor. Patient demand is another factor
influencing prescribing,(18) but there is only limited evidence to suggest that
patients prefer the buccal formulations."(19) (20) She paused.
"What about
those young whipper-snapper drug reps?" spluttered Colonel Mustard.
"That's true, those
young men can be very energetic..." said Miss Scarlett, thoughtfully. "I suppose that the
activity of drug representatives is a possible source of confounding."
"But the same
representative covers the whole of Buccalshire!" said Dr Watson.
"In addition, it's not
possible to tell from PACT data whether treatment was started by a hospital clinician or the
general practitioner-that is, whether a prescription was hospital initiated or not. This is
another potential confounder," said Miss Scarlett.
"I find all this a bit difficult to
digest," ruminated Dr Watson.
"It's alimentary, my dear Watson!" Colonel Mustard
exclaimed. "The most important variable in predicting glyceryl trinitrate buccal tablet
prescribing is the influence of secondary care, and this seems to be confirmed by the drug
company sales data" (table 4).
"Perhaps you should eat more slowly?" suggested Miss
Scarlett helpfully.
Epilogue
Miss Scarlett, Colonel Mustard, and Dr Watson set out
to discover why the prescribing of glyceryl trinitrate buccal tablets in Buccalshire was so
high. The most significant factor seemed to be the influence of prescribing at the interface
between secondary and primary care, in particular, the effect of hospital X.
Previous
work has shown that the buccal formulation of nitroglycerin is as effective as the
intravenous formulation of isosorbide dinitrate,(21) and consequently use of the
buccal formulation in hospitals is a sensible economic choice. Thus, this was probably not
an attempt to cost shift prescribing to primary care, but it may be a case of general
practitioners "learning by demonstration."(13) Whatever the cause of this "hospital
effect," the result on primary care is a significant cost burden on Buccalshire Family
Health Services Authority.
We thank Dr John Mucklow, Professor Peter
Jones, Dr Sean Fradgley, Pharmax, and the relevant people at hospitals X, Y, Z, and W.
Apologies to Waddingtons, manufacturers of Cluedo, and to Sir Arthur Conan Doyle, whose
character names we have used without any permission whatsoever.
| Key messages |
| The influence of secondary care establishments on prescribing in primary care has not been fully measured
Using glyceryl trinitrate buccal tablets as an example, this study measured the "hospital effect" in different catchment areas in a family health services authority
Catchment area was the single most important variable in predicting prescribing levels
A statistical model comprising catchment area, urbanity, number of partners in a practice, and a deprivation score explained over 47% of the variation in the prescribing of gyceryl trinitrate buccal tablets |
Funding: None.
Conflict of interest: None.
Department of Medicines Management,
Keele University,
Keele,
Staffordshire ST5 5BG
A J Pryce,
research assistant
H F Heatlie,
research assistant
S R Chapman,
director of prescribing analysis
Correspondence to: Ms A
J Pryce
Department of Mathematics,
MacKay Building,
Keele University,
Keele,
Staffordshire
ST5 5BG.
References:
1 World Health Organisation. Guidelines for
DDD. Oslo: WHO Collaborating Centre for Drug Statistics Methodology and the Nordic
Council on Medicines, 1991.
2 Edwards C, Metcalfe D, Burr A, Watson K, Steward F C N,
Jepson M H, et al. Influence of patient age on drug costs: an investigation to
validate the prescribing unit. Int J Pharm Pract 1991;1:73-8.
3 Reichek
N, Priest C, Kienzle M, Kleaveland JP, Bolton S, Sutton MStJ. Angina prophylaxis with buccal
synchron nitroglycerin: a rapid onset long-acting nitrate. In: Bussman W D, Dries R R, Wagner
W, eds. Advances in pharmacotherapy. Vol 1. Basle: Karger, 1982:143-54.
4 Abrams
J. Nitrate delivery systems-a contemporary perspective. Royal Society of Medicine
International Congress and Symposium Series 1983 No 54:13-24.
5 Roberts S J, Harris
C M. Age, sex, and temporary resident originated prescribing units (ASTRO-PUs): new
weightings for analysing prescribing of general practices in England. BMJ
1993;307:485-8.
6 Lloyd D C E F, Harris C M, Roberts D J. Specific therapeutic group
age-sex related prescribing units (STAR-PUs): weightings for analysing general practices'
prescribing in England. BMJ 1995;311:991-4.
7 Office of Population
Censuses and Surveys. Morbidity statistics from general practice. Fourth national study,
1991-1992. London: HMSO, 1995.
8 Wilson R P, Buchan I, Walley T. Alterations in
prescribing by general practitioner fundholders: an observational study. BMJ
1995;311:1347-50.
9 McCarthy M, Wilson-Davis K, McGavock H. Relationship
between the number of partners in a general practice and the number of different drugs
prescribed by that practice. Br J Gen Pract 1992;42:10-2.
10 Baker R,
Thompson J. Innovation in general practice: is the gap between training and non-training
practices getting wider? Br J Gen Pract 1995;45:297-300.
11 Morton-Jones
T, Pringle M. Prescribing costs in dispensing practices. BMJ 1993;306:1244-6.
12 NHS Executive. Purchasing and prescribing. Leeds: Department of Health, 1994:
paragraph 2. (EL(94)72.)
13 Strickland-Hodge B, Jepson M H. The role of the hospital
consultant in general practice prescribing. J R Soc Med 1988;81:207-9.
14
Healy A T, Yule B F, Reid JP. Variations in general practice prescribing costs and
implications for budget setting. Health Economics 1994;3:47-56.
15
Townsend P, Phillimore P, Beattie A. Health and deprivation. Inequality and the
North. London: Croom Helm, 1988.
16 Scrivener G, Lloyd D. Allocating census data to
general practice populations: implications for study of prescribing variation at practice
level. BMJ 1995;311:163-5.
17 Morris R, Carstairs V. Which deprivation? A
comparison of selected deprivation indexes. J Public Health Med
1991;13:318-26.
18 Audit Commission. A prescription for improvement. Towards
more rational prescribing in general practice. London: HMSO, 1994. (Health and personal
social services report, No 1.)
19 Bray C L, Jain S, Faragher E B, Myers A, Myers P,
MacIntyre P, et al. A comparison of buccal nitroglycerin and sublingual nitroglycerin
in the prophylaxis and treatment of exertional (situation-provoked) angina pectoris. Eur
Heart J 1991;12:16-20.
20 Ryden L, Schaffrath R. Buccal versus sublingual
nitroglycerin administration in the treatment of angina pectoris: a multicentre study.
Eur Heart J 1987;8:995-1001.
21 Lahiri A, Bowles M J, Brigden G,
Whittington J R, Duffy S W, Raftery E B. Buccal nitroglycerin and intravenous isosorbide
dinitrate in unstable angina. American Journal of Noninvasive Cardiology
1989;3:281-9.
