Received: 13 January 2020 Revised: 6 April 2020 Accepted: 5 May 2020

DOI: 10.1002/emp2.12116

ORIGINAL RESEARCH
Toxicology

Use of the online poisons information database TOXBASE

and admissions rates for poisoned patients from emergency
departments in England and Wales during 2008 to 2015

Kate Pyper PhD1 Chris Robertson PhD1,2 Michael Eddleston ScD3,4

Euan Sandilands MD4 D. Nicholas Bateman MD3
1
Department of Mathematics and Statistics, University of Strathclyde, Glasgow, UK
2
Health Protection Scotland, Glasgow, UK
3
Pharmacology, Toxicology & Therapeutics, University/BHFCentre for Cardiovascular Research, University of Edinburgh, Edinburgh, UK
4
National Poisons Information Service (Edinburgh Unit), Royal Infirmary of Edinburgh, Edinburgh, UK

Correspondence
Kate Pyper, PhD, Department of Mathemat- Abstract
ics and Statistics, University of Strathclyde,
Livingstone Tower, 26 Richmond Street,
Background: The impact of poison information services on patient care in hospital,
Glasgow G1 1XH, UK. particularly decisions on whether to admit patients after initial attendance at an emer-
Email: [email protected]
gency department (ED), is unclear. In the United Kingdom, the vast majority of poisons
Funding and support: By JACEP Open policy, information is provided by use of the online poisons information database, TOXBASE.
all authors are required to disclose any and all
We investigated the relationship between rates of hospital access to TOXBASE and
commercial, financial, and other relationships
in any way related to the subject of this article rates of poisoning admissions from EDs in England and Wales to begin to address the
as per ICMJE conflict of interest guidelines (see
interactions between use of poisons information and patient management as reflected
www.icmje.org). The authors have stated that
no such relationships exist. by hospital activity.
Methods: Data were obtained on attendances and admissions due to poisoning for
individual National Health Service (NHS) Trusts in both England and Wales, together
with data on the overall number of accesses to TOXBASE for drugs (pharmaceuticals
and drugs of abuse), from 2008 to 2015. Rates of TOXBASE access and admissions
per poisoning attendance in London were clearly different to the rest of England and
Wales; London was therefore analyzed separately. Negative binomial generalized addi-
tive models were fit, incorporating an interaction effect, for accesses, attendances and
admissions to check for variability according to hospital size. Additional models were
then fit to assess whether there was any variation in association of overall TOXBASE
use with rates of admission for 6 key drug subgroups: antidepressants, paracetamol,
antipsychotics, opioids (including all medicines, but excluding heroin), heroin and non-
opioid drugs of abuse.

Supervising Editor: Christian A. Tomaszewski, MD, MS.

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any
medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2020 The Authors. JACEP Open published by Wiley Periodicals LLC on behalf of the American College of Emergency Physicians.

JACEP Open 2020;1–12. wileyonlinelibrary.com/journal/emp2 1

2 PYPER ET AL .

Results: Rates of TOXBASE use per Trust increased across the study period by 39.3%
(95% confidence interval [CI] = 34.1%, 44.8%) in England and 76.9% (24.7%, 151.0%) in
Wales, showing an increase in TOXBASE use which was substantially greater than the
increase in poisoning attendances. Admission rates exhibited seasonality, with lower
rates in January and February, increasing by 2.0% (1.0%, 3.1%) in England and 5.8%
(5.5%, 5.9%) in Wales toward the middle of the year. The initial model fit indicated that
the average proportion of poisoning patients admitted increased with both increas-
ing attendances and increasing TOXBASE use (England and Wales overall, P < 0.0001;
England and Wales excluding London, P < 0.0001; London, P < 0.0001). In England and
Wales overall, and in London alone, increased TOXBASE access to non-opioid drugs
of abuse advice was associated with a significant decrease in admissions (England and
Wales, −0.15% [−0.29%, −0.01%] [P = 0.032]; London, −1.02% [−1.53%, −0.50%]
[P < 0.0001]). In contrast, increased access to heroin advice was associated with a
significant increase in admissions in London (+2.03% [+0.11%, +3.99%] [P = 0.034]).
Increasing access to TOXBASE for paracetamol advice was associated with lower
admissions in England and Wales (England and Wales, −0.11% [−0.23%, −0.01%]
[P = 0.036]; England and Wales excluding London, −0.18% [−0.30%, −0.06%] [P =
0.001]) but higher admissions in London (+0.52% [+0.03%, +1.01%] [P = 0.035]).
Conclusions: We have shown that greater overall use of TOXBASE by hospitals is asso-
ciated with a higher proportion of poisoning attendances being admitted. Interestingly,
looking at particular drug groups, we found significant associations in both directions
between overall TOXBASE use and rates of admission for some drug groups. The cur-
rent methodology is unable to determine whether such decisions might be appropriate
or not. Mixed-methods research is now required to gain a better understanding of how
provision of poisons information affects decisions within the ED.

KEYWORDS
drugs of abuse, hospital activity analysis, pharmaceuticals, poisons information

1 INTRODUCTION mary and secondary care services.3 The UK National Poisons Informa-
tion Service had launched its online clinical database TOXBASE in late
1.1 Background 1999; it was subsequently adopted as the primary source of poisons
information for all healthcare services across the United Kingdom.4,5
Poisons information services aim to improve the care of patients with TOXBASE is used routinely in the management of poisoned patients,
suspected or confirmed poisoning. They may prevent unnecessary hos- providing information, not only on the potential toxicity and treat-
pital presentations (attendances) and are intended to improve triage ment of compounds, but also the dose thresholds for toxic effects and
(decision whether to admit the patient to hospital or discharge from required duration and nature of clinical monitoring. The database is
the emergency department) and care of patients who do present to written and regularly updated by the National Poisons Information
hospital. When used by the public or primary care services, provi- Service clinical staff and has an editorial team with specialist advi-
sion of poisons information reduces attendances at emergency depart- sors and based on national and international published advice. It can
ments and hospital admissions in both North America and the United be searched for generic and common brand name products, particu-
Kingdom.1,2 larly medications. It also contains data on all types of poisoning. It is
In the United Kingdom, the public is unable to access poisons infor- the agreed standard of care in the United Kingdom. Where additional
mation directly from poisons information centers. The National Health information is required, healthcare workers call a single national phone
Service’s (NHS) introduction of telephone health information lines for number to be connected with poison information specialists and clini-
the general public in the early 2000s led to the need for a single cal toxicologists at 1 of 4 National Poisons Information Service poisons
authoritative approach to poisons information provision for both pri- information centers.
PYPER ET AL . 3

Since 1999, use of TOXBASE has increased by >600% and contin-

ues to increase year by year. In 2017–2018, there were over 625,000 The Bottom Line
individual log-ons in the United Kingdom to the database, access-
Use of poison information services in emergency depart-
ing in excess of 1.6 million product (agent) entries. Most (60%) came
ments may impact admission decisions. Although greater
from hospitals, although 30% came from primary care and ambu-
TOXBASE use in England and Wales was associated with
lance services.6 Most hospital enquiries come from emergency depart-
admission for poisoning overall, there was a decrease in
ments; TOXBASE accesses account for the vast majority of National
admission for non-opioid drugs of abuse and acetaminophen.
Poisons Information Service contacts from such departments (354,678
in 2016–2017 compared to <11,000 telephone enquiries). For this rea-
son, we concentrated our analysis on TOXBASE use.

2 METHODS
1.2 Importance
2.1 Design
Exploring how poisons information services impact patient care in hos-
pital is challenging. Assessing how poisons information affects triage Data on the number of admissions and attendances for poisoning in
and handling within hospital emergency departments themselves is an England and Wales for the period January 2008 to December 2015
important step in understanding, and auditing this interaction is key to were obtained from the Health and Social Care Information Centre
improving patient care and the use of poisons services in acute hos- (HSCIC; https://digital.nhs.uk/) and from the NHS Wales Informatics
pitals. This is an important issue if poisons services are to be used Service (http://www.wales.nhs.uk/nwis/home). These data were then
optimally and developed to respond to the needs of emergency physi- linked with data on the number of accesses made to the TOXBASE
cians. This project was undertaken as a first step in understanding how database. For Wales, this was a straightforward task of linking by hospi-
current services are used, and how this might vary across the United tal name. The linkage for the English data were more difficult, because
Kingdom. in some cases the hospital activity data were provided at individual
hospital level and in others it was provided at the NHS trust level
(an administrative unit), amalgamating data from 2 or more hospitals.
1.3 Goals For consistency, all English hospital-level data were aggregated up to
NHS Trust level, which were linked with the relevant hospitals in the
The aim of this project was to evaluate the impact of poisons infor- TOXBASE data set.
mation services on physician behavior. This is complicated, because
hospital activity varies both seasonally and geographically. In addition,
use of the TOXBASE database has increased over time.7 In England 2.2 Setting
and Wales (England and Wales), routine data are available on atten-
dances at emergency departments, with the broad clinical area of rea- This study was conducted using data available from routine NHS
son for attendance described (eg, poisoning), and the proportion of data sources on attendance and admissions from hospital emergency
these attendances that are admitted.8,9 Hospital admissions activity is departments in the United Kingdom. We also assessed overall usage
coded using ICD10, but this coding is not yet applied to attendances. of TOXBASE pharmaceutical product (agent) entries (including drugs
In this study, we used routinely collected data on TOXBASE accesses of abuse) as well as 6 common medicines and recreational drug groups:
from individual NHS Trusts (hospitals) to these pharmaceutical (drug) antidepressants, paracetamol, antipsychotics, opioids (including all opi-
products on TOXBASE and compared these to attendances, and admis- oid medicines but excluding heroin), heroin, and non-opioid drugs of
sions to wards from the ED, from 2008–2015. Most TOXBASE accesses abuse. For simplicity, we refer to these together as “Drugs.” Heroin was
(83%) from hospitals relate to pharmaceutical (drug) products, includ- assessed separately from opioids due to the major differences in its
ing drugs of abuse. Using these data, we have explored the relationship supply and usage across the United Kingdom.
between frequency of overall use of TOXBASE and changes in rates
of admission as a proportion of attendances. In this way, we hoped to
begin to understand how the use of online poisons information ser- 2.3 Selection
vices in the United Kingdom might interact with clinician behavior as
reflected by hospital activity and admissions in order to design future The data set used contained all data available to us for the study period
studies. An epidemiological study of this design shows associations, in England and Wales. Because poisoning attendances were not con-
which can then be the focus for further studies; it cannot establish cau- sistently recorded in the early days of these NHS data systems, some
sation. As online information systems are increasingly used in medicine data were excluded from these analyses. Attendance reasons were not
this project may also inform others in the field. recorded in Wales until April 2009, so all Welsh data prior to that
4 PYPER ET AL .

date were eliminated from the analysis. In England, the reasons for In this model, E(log (yi )) is the expected log number of admissions,
attendance at EDs (ie, poisoning) were not always consistently and log (ai ) is the log number of poisoning attendances. The result
recorded in some Trusts, and this was further complicated by merger of this is that this model provides insight into the rate of poisoning
of hospitals into Trusts within the study period. Together, this resulted admissions per poisoning attendance, rather than the number of admis-
in some large data gaps for poisoning attendances and resultant sharp sions due to poisoning. We wanted to incorporate some measure of
drops in the number of poisoning attendances recorded. Data within hospital size in the model, hence the total number of attendances is
these periods of unreliable recording were eliminated on a case-by- represented by x1i . The rate of TOXBASE access per attendance is
case basis resulting in 4986 rows of data out of 13,990 rows being represented in the model using x2i , and x1i x2i denotes an interaction
omitted from the analysis; this amounted to only 2.8% of the total between hospital size and TOXBASE accesses, which will account for
number of attendances at A&E for poisoning (1,255,412). As a con- differences in the effect of TOXBASE access by hospital size. The final
sequence, data on 37.3% of TOXBASE accesses and 37.4% of hospi- 2 terms represent the overall (f(ti )) and seasonal (g(mi )) patterns in the
tal admissions could not be used indicating that the data quality issue rate of admissions per attendance. Positive values of β1 or β2 indicated
only affected emergency attendances to hospital. In total, the analysis that admissions grow with attendances or accesses, whereas negative
included data from 116 trusts out of 143 in England and 13 out of 17 values indicated a decrease in admissions with attendances or accesses
in Wales. respectively. The coefficient β3 of the interaction term indicated how
Data obtained from HSCIC in England were subject to the sup- admissions change with attendances for different levels of TOXBASE
pression of small values (between 1 and 5) to keep individual patients usage.
unidentifiable. To address this issue, we used a technique whereby The above model was used as a starting point for the analysis
missing values were imputed empirically using a Bayesian model incor- and was simplified where appropriate. These data exhibited over-
porating seasonality and long-term trend to estimate results in the dispersion, a common artefact in count data, where the variance is
range of 1–5. Use of this method made allowance for hospital size (and larger than the mean. This does not hold with the assumption of equal
therefore workload) that meant that missing values for those hospi- mean and variance distribution made by use of the Poisson distribution.
tals that were larger and consequently had mostly complete data were In order to account for this over-dispersion, a negative binomial model
more likely to be imputed as 4 or 5, whereas those locations with a was used.
large number of missing values were likely to be smaller hospitals and Similar models were then used to summarize how the rate of admis-
so more likely to have smaller imputed values. sion varied in an average trust, with respect to the proportion of
This study did not require approval by a UK Research Ethics Com- accesses to the 6 aforementioned drug sub-groups. These models took
mittee because the UK Health Research Authority has declared that a simpler form as outlined in the following equation:
ethical approval is not needed for research studies that use information
collected routinely by any UK administration (England, Wales, Scotland E(log(yi )) = 𝛽0 + 𝛽1 xi . (2)
and Northern Ireland) as part of usual clinical care, provided this infor-
mation is passed to the researchers in a fully anonymized format. Here, yi is the proportion of attendances due to poisoning that
resulted in admission within each Trust, whereas xi is the proportion
of accesses made to the relevant drug group. Each drug group was
2.4 Analysis assessed independently, but the results will be presented as a whole. In
this analysis, we expect the coefficient β1 to be negative when increas-
The seasonal and long-term trends in the rate of poisoning admis- ing proportions of accesses are associated with a decreasing propor-
sions and the rate of TOXBASE accesses were explored using gener- tion of admissions.
alized additive models, which allow for smoothly varying relationships Temporal changes were explored by fitting models for rates of
through time. The smooth terms and parameters in these models were TOXBASE access and rates of admission by trust, through time. The
estimated via an iterative process. At each step, the terms were esti- models were of the form:
mated based on the response values minus the current estimates of
each other term in the model, so that the values being used to estimate E(log(y)) = 𝛽0 + 𝛽1 year. (3)
the components were overall centered around zero but still retained
some pattern that is to be estimated, herein referred to as the centered where y represents either the rate of access to TOXBASE or the rate
response. It is the final values of this iterative process that will be pre- of admission due to drugs poisoning. The coefficient β1 was extracted
sented and interpreted as the trends in the model. for admissions and accesses for each trust to give a measure of
The first model in this article examined the appropriateness of an the rate of change of TOXBASE use and admissions over time by
interaction term in examining the effect of TOXBASE use on admis- trust.
sions. The model fit was of the form shown below: Due to the nature of the data, each model was fit on the log
scale, meaning that differences estimated from any of the above mod-
E(log(yi )) = log(ai ) + 𝛽0 + 𝛽1 x1i + 𝛽2 x2i + 𝛽3 x1i x2i + f(ti ) + g(mi ). (1) els are presented as multiplicative percentage changes. The results
PYPER ET AL . 5

England Wales

Centred Response
A. Long−Term Trend B. Long−Term Trend

Centred Response

0.2 0.4
0.0
−0.2

−0.2
2008 2010 2012 2014 2016 2009 2011 2013 2015

Date Date

C. Seasonal Trend D. Seasonal Trend

0.02
Centred Response

Centred Response

0.05
0.00

−0.05
−0.02

Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov

Month Month

F I G U R E 1 Plots showing the long-term and seasonal trends for TOXBASE accesses in grey (―) and poisoning admissions in black (―) for
England (A and C) and Wales (B and D). The dashed lines represent 95% confidence intervals. For ease of comparison the data are shown related to
a centered response

section will also describe raw data, estimated regression coefficients TOXBASE accesses (88.9% increase; 2.37/attendance). In Wales, the
and model predictions where appropriate. respective figures for 2009 were 425 attendances, 207 admissions
(48.7% of attendances) and 686 TOXBASE accesses (1.61/attendance);
by 2015, these had risen to 641 (50.8% increase), 291 (40.6% increase;
3 RESULTS 45.4% of attendances) and 2239 (326% increase; 3.49/attendance),
respectively.
During the period studied, after accounting for issues with data record- Using the statistical models to look at the overall trend, the pro-
ing, there were 1,220,857 attendances at hospital emergency depart- portion of attendances admitted decreased by 17.0% (14.3%, 19.6%)
ments in England and Wales that were coded as due to poisoning. Dur- in England between 2008 and 2015 compared to the non-significant
ing the same period, 581,368 patients were admitted (46.5 per 100 decrease of 4.9% (−23.0%, 13.2%) in Wales between 2009 and 2015.
attendances) due to overdose involving Drugs, and the relevant prod- TOXBASE usage, as a rate per poisoning attendance, in the statistical
ucts pages within TOXBASE were consulted 2,629,846 times (2.15 models increased over the study period, by 39.3% (34.1%, 44.8%) in
accesses/attendance). England and 76.9% (24.7%, 151.0%) in Wales.
We assessed whether the admission rate and the rate of TOXBASE
accesses per poisoning attendance showed seasonal variability. The
3.1 Long-term and seasonal trends proportion of attendances admitted showed statistically clear sea-
sonal variability, with a peak in May in England and August in Wales
The seasonal and long-term trends in the rate of admissions due to (Figure 1); the seasonal peak in summer is equivalent to a small increase
drugs poisoning and the rate of TOXBASE accesses/month are shown in admissions of 2.0% (1.0%, 3.1%) in England, and 5.8% (5.5%, 5.9%)
in Figure 1 as a centered response. In England, the average Trust in Wales from January. Seasonal variability also occurred in the vol-
had 1262 attendances, 639 admissions (50.6% of attendances), and ume of TOXBASE use as we have previously reported7 ; however, exam-
2205 TOXBASE accesses (1.75/attendance) in 2008. In 2015, the aver- ining TOXBASE accesses per poisoning attendance to control for the
age number of attendances had increased to 1754 (39.0% increase), seasonal variability in attendances shows that there is no residual sea-
with 762 admissions (19.2% increase; 43.4% of attendances) and 4165 sonality in TOXBASE access.
6 PYPER ET AL .

TA B L E 1 Total number of attendances, poisoning-related admissions and TOXBASE accesses, in addition to the proportion of attendances
admitted and the rate of TOXBASE access per attendance for each region between 2008 and 2015

Poisoning- Rate of
Related TOXBASE TOXBASE
Attendances Admissions Proportion of accesses access per
Region (n) (n) admissions (n) attendance
North West 241,883 119,105 0.492 553,592 2.289
Yorkshire and the Humber 183,263 82,582 0.451 387,800 2.116
South East 163,955 84,843 0.517 328,887 2.006
London 150,815 42,808 0.284 246,445 1.634
West Midlands 124,769 57,051 0.457 239,882 1.923
East of England 95,617 53,290 0.557 250,536 2.62
South West 82,753 45,160 0.546 168,682 2.038
East Midlands 73,586 41,455 0.563 189,344 2.573
North East 56,303 32,460 0.577 130,550 2.319
Wales 47,913 22,614 0.472 134,128 2.799
Total 1,220,857 581,368 0.476 2,629,846 2.154

F I G U R E 2 "Heat" maps showing the variation in the proportion of poisoning attendances admitted and the rate of TOXBASE accesses per
poisoning attendance. Darker areas represent higher activity

3.2 Regional variability We found there to be little variability in the proportion of poisoning
attendances who were admitted, other than for London. Over the
We then investigated regional variability in the proportion of poison- entire study period, the proportion of attendances admitted from Eng-
ing attendances admitted and in the rate of TOXBASE accesses per land and Wales regions outside London ranged from 0.451 in Yorkshire
poisoning attendance across England and Wales. This was done by and the Humber region to 0.577 in the North East region (Table 1;
examining NHS Regional level maps of the data. These NHS Regions Figure 2) compared to 0.284 in London. The rate of TOXBASE accesses
were changed during the period covered by the study to a much per poisoning attendance within England varied from 1.923 in the
larger number of smaller areas run by NHS Clinical Commissioning West Midlands to 2.620 in the East of England, with London again low
Groups. For simplicity, we have used the older regional classification. at 1.634 and Wales high at 2.799 (Table 1; Figure 2).
Overall rates of admission and TOXBASE access per attendance were From these data, it is clear that the pattern of poisoning admissions
calculated as the total number of admissions or accesses in that region, and TOXBASE accesses is different in London compared to the rest
over the entire period, divided by the total number of attendances due of England and Wales. Specifically, both the proportion of attendances
to poisoning in that region over the entire period. admitted and the rate of TOXBASE accesses per poisoning attendance
PYPER ET AL . 7

TA B L E 2 Model coefficients for the effect of hospital attendances, TOXBASE accesses and the interaction of these attendances and accesses
on hospital admission due to drugs poisoning for each of England and Wales overall, England and Wales excluding London and London only. These
coefficients were estimated after accounting for temporal trends in admissions

England and Wales England and Wales

overall excluding London London
Attendances −2.570 × 10−5 −2.069 × 10−5 −3.453 × 10−5
SE 1.57 × 10−6 SE 1.92 × 10−6 SE 3.55 × 10−6
P < 0.0001 P < 0.0001 P < 0.0001
TOXBASE accesses 0.1107 0.1198 0.0565
SE 4.48 × 10−3 SE 4.58 × 10−3 SE 2.04 × 10−2
P < 0.0001 P < 0.0001 P < 0.0001
Attendances × TOXBASE 4.073 × 10−6 2.931 × 10−6 8.405 × 10−6
accesses SE 4.52 × 10−7 SE 4.76 × 10−7 SE 1.77 × 10−6
P < 0.0001 P < 0.0001 P < 0.0001

TA B L E 3 Estimated coefficients for the effect of hospital size (as measured by the total number of attendances) and the rate of access to
TOXBASE on the rate of admission due to poisoning per poisoning attendance after accounting for temporal trends

England and
England and Wales excluding
Wales overall London London
−5 −5
Total attendances −1.719 × 10 −1.354 × 10 −2.185 × 10−5
SE 1.28 × 10−6 SE 1.55 × 10−6 SE 2.49 × 10−6
P < 0.0001 P < 0.0001 P < 0.0001
Rate of access to TOXBASE 0.1447 0.1435 0.1438
SE 0.0025 SE 2.58 × 10−3 SE 9.14 × 10−3
P < 0.0001 P < 0.0001 P < 0.0001

are lower in London than in the other regions. One possible explana- For the data overall, the model including the interaction term
tion for the difference in London may be relative excess of medical stu- resulted in 75.4% of the variance in the observed data being
dents, but this would be expected to cause more TOXBASE accesses explained—this was only a marginal improvement over the model
per admission, not the fewer we saw. This observed difference may also excluding the interaction term, which had 75.2% variance explained.
be related to some London hospitals following different protocols for These results were similar for the data excluding London and for Lon-
treating poisoned patients, and this prompted the decision to further don individually. This indicates that the interaction effect does not con-
assess trends in 3 different ways: England and Wales overall, in England tribute a great deal to the fit of the model, and for simplicity, the model
and Wales without London and London alone. excluding this interaction will be presented here.
We found that the rate of admissions due to poisoning (per poison-
ing attendance) tended to decrease as hospital size increased, as indi-
3.3 Interaction models cated by the negative estimate of the coefficient in the first row of
Table 3. This decrease is such that for every 100 additional attendances
We then assessed the relationship between rate of TOXBASE use the rate of admissions per poisoning attendance would decrease by
per poisoning attendance and the rate of admissions due to drugs ≈0.2%.
poisoning per poisoning attendance. This was initially done using the In addition, there was an increase in the rate of poisoning admis-
interaction model outlined in Equation 1, which took account of the sions with an increasing rate of access to TOXBASE. This increase
effect of overall attendances on the rate of admissions, to investigate was such that for every additional TOXBASE page accessed per poi-
whether the effect of TOXBASE use varied depending on hospital size. soning patient, there was a 15.6% (95% confidence interval [CI] =
We found that the interaction term in the model was 4.073 × 10−6 15.0%, 16.1%) increase in the rate of poisoning admissions per poison-
(standard error (se) 4.515 × 10−7 ), P < 0.001. This implies that were ing attendance. The size of this effect was similar across all 3 of the
large hospitals to increase their rate of accesses to TOXBASE the pre- regional configurations examined (see Table 2; positive coefficient in
dicted increase in the admission rate is proportionately higher than in the second row).
a smaller hospital which increases its access rate. However, the mag- These models were additionally fit for each year to investigate
nitude of this effect is minor as evidenced by the estimated coeffi- whether the relationship between TOXBASE use and attendances was
cient, given the magnitude of the coefficient of the main effect of the stable over the period of the study. With the exception of 2008, dur-
TOXBASE access rate (Table 2). ing which there were no observations for Welsh hospitals, and English
8 PYPER ET AL .

Rate of Admission due to Poisoning

0.50
0.45
2008
2008
2009
2009
2010
2010
2011
2011
2012
2012
0.40

2013
2013
2014
2014
2015
2015

1.6 1.8 2.0 2.2 2.4 2.6

Rate of Access to TOXBASE

F I G U R E 3 Plot showing the relationship between the rate of poisoning admissions per poisoning attendance and the rate of TOXBASE use per
poisoning attendance. Groups of points are separated by colour to extract the relationship by month within each year

data not particularly well recorded, there was very little variation in the and the largest decrease in TOXBASE use were (coefficient [se]): 0.049
estimated coefficients by year (Supporting Information Table S1). (0.020), 0.109 (0.022) and 0.203 (0.027), respectively. This indicates
The positive association between TOXBASE use and admissions that for those hospitals in which TOXBASE use has decreased most,
described by the model coefficients seems counterintuitive, given the the rate of admissions reduced more than the comparative increase in
opposing trends in the rate of admission and the rate of TOXBASE admissions in those hospitals whose TOXBASE use increased most, fur-
use seen in Figure 1. Further exploration of this is shown in Figure 3, ther emphasising an interaction between TOXBASE use and admission
which plots the rate of admissions per poisoning attendance against rates with poisoning.
the rate of TOXBASE access per poisoning attendance. The data have This finding is suggestive of a link between TOXBASE use and triage
been aggregated over trusts, so that each point in the plot corresponds admission decisions for poisoned patients. However, this analysis can-
to a month of data. The points are colored based on which year the not determine whether TOXBASE use is changing decisions on admis-
data were observed in, and it can be seen that points observed in sion or whether more severe cases that require admission result in
earlier years tend to be on the left-hand side of the plot, with later more TOXBASE use. To explore this aspect in more detail, we under-
years sitting on the right-hand side of the plot. There is, overall, a took modelling of common drug groups, to compare the patterns of
decreasing trend in these observations, which is due to the overall TOXBASE agent accesses for these drugs with the overall proportion of
temporal pattern, where the average rate of admission has decreased poisoning patients admitted in NHS Trusts, which might lead to a better
over time. understanding of the impact of TOXBASE. We reasoned that, if these
Examining individual groups of points, it can be seen that, particu- drug groups varied in their effects on admission, this might indicate an
larly in recent years, there is positive correlation between the rate of effect deriving from TOXBASE advice.
admission and the rate of TOXBASE access per poisoning attendance.
It is this which is driving the positive coefficient for TOXBASE use in the
models above (Tables 2 and 3). 3.4 Drug group specific modelling
Models were fit for each individual trust for the rate of TOXBASE
accesses and the rate of admissions due to drug poisoning against year We did not have access to diagnostic coding for attendances or admis-
to get an estimate of the annual rate of change. These trust-level coef- sions, preventing us from relating drug-specific TOXBASE accesses to
ficients were used to find those trusts that had the largest decrease attendances and admissions for that same drug. However, we obtained
in TOXBASE use through time, the largest increase in TOXBASE use drug group TOXBASE access data for 6 important drug groups (antide-
through time, and the smallest change in TOXBASE use through time. pressants, paracetamol, antipsychotics, opioid medicines [excluding
The top 5 in each case were taken, and the rate of admission due to poi- heroin], heroin, and non-opioid drugs of abuse, Table 4) and com-
soning was plotted against the rate of TOXBASE access per poisoning pared it with the overall proportion of poisoning admissions per
attendance. The coefficients for the relationship between admission Trust. Across groups, we noted regional variation in TOXBASE
and access rates in those with the largest increase, the smallest change, accesses, with London often being high or low for the proportion
PYPER ET AL . 9

TA B L E 4 Number of accesses to TOXBASE entries on pharmaceuticals overall and to selected drug classes during the study period by UK areas

England and England and Wales

Walesoverall(%) excluding London(%) London(%)
All drug accesses 2,184,507 1,985,033 199,474
Antidepressants 370,657 (17.0) 341,922 (17.2) 28,735 (14.4)
Paracetamol 323,308 (14.8) 294,449 (14.8) 28,859 (14.5)
Non-opioid drugs of abuse 150,040 (6.87) 133,252 (6.71) 16,788 (8.42)
Opioids 141,069 (6.45) 130,612 (6.58) 10,457 (5.24)
Antipsychotics 125,698 (5.75) 113,912 (5.74) 11,786 (5.91)
Heroin 15,786 (0.72) 13,992 (0.70) 1,794 (0.90)

Each cell contains the number of accesses. The percentage of accesses is displayed in parentheses where appropriate.

TA B L E 5 Individual effects of accesses to TOXBASE on the rate of change in admissions as presentations for 6 common drug sub-groups for
England and Wales combined, England and Wales excluding London, and London trusts only

England and England and Wales

Walesoverall excluding London London
Antidepressants +0.01% +0.04% −0.21%
(−0.11%, +0.14%) (−0.09%, +0.17%) (−0.71%, +0.30%)
Paracetamol −0.11% −0.18% +0.52%
(−0.23%, −0.01%)a (−0.30%, −0.06%)a (+0.03%, +1.01%)a
Non-opioid drugs of abuse −0.15% −0.04% −1.02%
(−0.29%, −0.01%)a (−0.19%, +0.10%) (−1.53%, −0.50%)a
Antipsychotics +0.00% +0.07% −0.49%
(−0.19%, +0.20%) (−0.14%, +0.28%) (−1.21%, +0.24%)
Opioids −0.02% +0.03% −0.59%
(−0.22%, +0.17%) (−0.17%, +0.23%) (−1.41%, +0.23%)
Heroin +0.44% +0.16% +2.03%
(−0.15%, +1.03%) (−0.45%, +0.78%) (+0.11%, +3.99%)a

The values shown are the percentage change in admissions for every increase of 1 attendance. a Significant effects.

of accesses to each drug group (Table 4, Supporting Information 4 DISCUSSION

Table S1).
Models were fit to assess the effect of TOXBASE access rates on In this paper, we examined the interaction between hospital activity—
admissions for these drug groups. The effects of the drugs categories specifically the decision to admit a poisoned patient attending the
on admission were in some cases quite different between London and emergency department—and the level of use of the online poison infor-
elsewhere in England and Wales (Table 5). In England and Wales over- mation service TOXBASE. We used statistical models for this analy-
all, and in London alone, increased accesses to TOXBASE advice on sis, to account for underlying national changes over time of the use of
non-opioid drugs of abuse (as a proportion of all TOXBASE accesses) the TOXBASE database, poisoning attendances, and admissions from
were associated with a statistically significant decrease in admissions the emergency department to hospital wards. The United Kingdom
per attendance (England and Wales, −0.15% [−0.29%, −0.01%], P = runs a health care system (NHS) that is free at pointofcare and is fully
0.032; London, −1.02% [−1.53%, −0.50%], P = 9.2 × 10−5 ). Increased funded by taxation. There are no private facilities included in this anal-
access to heroin advice was associated with a significant increase ysis and effectively none in the United Kingdom in any case. Patients
in admissions in London (+2.03% [+0.11%, +3.99%], P = 0.034) but may, as elsewhere, self-present to an ED, present after consultation
not elsewhere in England and Wales. Increased access to the parac- with a doctor/pharmacist or generic health information telephone
etamol TOXBASE entry was associated with a significant reduction line (NHS111 in UK) or after calling an ambulance directly. Although
in admissions in England and Wales overall and England and Wales these data do not allow us yet to understand the interaction of poi-
excluding London (England and Wales, −0.11% [−0.23%, −0.01%], P = sons information and management at an individual patient level, being
0.036; England and Wales excl. London, −0.18% [−0.30%, −0.06%], P = designed to give an overview of the interactions between database use
0.001) but increased admissions in London (+0.52% [+0.03%, +1.01%], and triage at a national, regional and drug category level across Eng-
P = 0.035). land and Wales, they are essential for describing and understanding
10 PYPER ET AL .

the situation before conducting any further mixed-methods research contrast, if TOXBASE influenced clinical decisionmaking, one would
analysis. expect the relationship between specific TOXBASE access and admis-
The use of an internet database by clinicians is not equivalent to sions to be more complex—the situation revealed in this analysis. Inter-
telephone enquiries, in that >1 access may be made to an individ- estingly we found both positive and negative relationships between
ual database drug entry for individual patients. Thus, access rates do admissions and TOXBASE accesses for different drug groups and in dif-
not equate to patients, and more complex poisoning cases may result ferent regions. Increased rates of access to non-opioid drugs of abuse
in more database accesses than simpler ones. It is not possible for were associated with reduced admissions, whereas increased access
us to differentiate duplicate enquiries from single enquiries about a to paracetamol information was associated with increased admissions
patient. Because the number of telephone calls from UK hospitals to in London but reduced admissions in the rest of England and Wales.
the National Poisons Information Service were orders of magnitude Thus, although it might be assumed that staff might be familiar with
less than TOXBASE use, only 3% of the National Poisons Information common poisoning such as paracetamol and only use TOXBASE for dif-
Service enquiries coming from hospital emergency departments are ficult patients, there are several points in the management pathway
by telephone, they are unlikely to impact this analysis. The approach where reference to TOXBASE might be useful, such as reviewing blood
we have used is intended to address this by overall modelling of total test results and decisions on further treatment or discharge. Increased
database activity, and patient handling is unlikely to be unique to access to heroin information was associated with an increase in admis-
individual hospitals or regions of the United Kingdom. We included sions in London only.
large numbers of TOXBASE enquiries and related them to hospital These findings are compatible with clinical decisionmaking being
activity. related to TOXBASE advice with the observed data being what one
These models show clear interactions on a population level between might expect if poisons advice aided triage of some patient groups,
the usage of poisons information services and hospital activity. For any enabling early safe discharge of some, and requiring further observa-
given level of activity (number of poisoning attendances), the number tion of others, with admission. Although we cannot definitively con-
of patients admitted directly relates to the number of times TOXBASE firm this due to lack of data on presentation medication ingestion, we
is accessed. Higher rates of TOXBASE access are associated with higher believe this the most plausible explanation.
rates of poisoned patient admissions, especially in larger hospitals. The data on paracetamol are unexpected and may relate to some
These interactions seemed stable across all years studied, apart from clinicians accessing TOXBASE for all paracetamol poisoning due to the
2008 when data from the Welsh system was unavailable and English complexity of advice since the 2012 regulator’s new advice on parac-
data markedly incomplete (Supporting Information Table S1). Compar- etamol management,10 and in particular use of local management pro-
ing hospitals based on their change in TOXBASE use over time indi- tocols, such as we have previously described.11
cated that, in the hospitals where the usage of TOXBASE decreased The patterns of attendances and TOXBASE use are different in Lon-
most over the study period, there was a larger relative decrease in the don compared to the rest of England and Wales. These findings are in
rate of admissions compared to the increase in admissions in those hos- line with the regional differences we have previously found in paraceta-
pitals whose TOXBASE use increased. These analyses further support mol poisoning admissions12 and with different patterns of use of drugs
interactions between TOXBASE use and admission rates as they show of abuse across the United Kingdom.13 London is the NHS Region with
clear relationships between hospital activity and changes and usage of least TOXBASE use and fewest admissions for poisoning. Although
poisons information in TOXBASE. the current data on hospital attendances are insufficiently detailed to
It is possible that higher TOXBASE use is due to more complex cases allow any further exploration of this finding, possible explanations may
being seen and admitted, and the database being accessed more often include the high proportion of English medical schools based in London
by multiple clinical teams involved in the treatment of these patients. (7 of 25 total) and a difference in experience or confidence of ED staff.
However, because this methodology does not determine causality, Other possibilities are differences in epidemiology of poisoning, local
we cannot ascertain if the increased admission rate per attendance practice or the way TOXBASE is used between London and elsewhere
associated with increased TOXBASE use represents improved patient in England and Wales. Further research will require more data on indi-
management or is caused by an unknown confounder. Although the vidual case-mix and severity, as well as direct observational studies of
increased use of a poison information facility could affect admissions, TOXBASE use. However, these differing patterns and trends within dif-
further work is required to assess if the increases in admissions we ferent UK regions are compatible with a direct effect of poisons infor-
observed are clinically appropriate. Unfortunately, national attendance mation on clinician behavior.
data are not coded accurately enough for us to account for the influ- Going forward, we believe these data have the potential to be used
ence of attendance case-mix on proportion of cases admitted. for audit of performance of individual units as a tool for public health,
We addressed this problem in attendance coding by reasoning that and to do that, optimally, we would need data on presentation inges-
analysis of TOXBASE accesses to groups of different pharmaceuticals tions to be included, which is not presently the case. This has also
could act as a surrogate for case-mix. If use of TOXBASE was entirely prevented analysis of admissions as a proportion of presentations for
driven by hospital activity, then the relationship between admissions specific drugs. This is one reason we are unable to assess the reasons
and ED accesses to relevant TOXBASE pages would be consistent. In for different patterns in London and the rest of the United Kingdom.
PYPER ET AL . 11

Such information would also better support policy in managing poison- Further mixed-methods research is required to obtain better evidence
ing in the United Kingdom. on how TOXBASE is used in EDs and how it might affect clinical
practice.

5 LIMITATIONS AUTHOR CONTRIBUTIONS

KP and CR formulated and carried out the analysis presented. ME and
The paper is limited by our inability to show causality. However, ES contributed data and advice relating to TOXBASE. DNB and KP
because we show that the rates of admission are related to TOXBASE drafted the manuscript and all authors contributed to its revision.
use at Trust level, and these effects are significant at the national level,
the association has face validity. CONFLICT OF INTEREST
Data had to be excluded from the analysis due to incompleteness or The authors declare no conflict of interest.
to inconsistency in the way attendance reasons were recorded. We are
dependent on hospital coding for our analyses, with discharge coding ORCID
being used in the case of admission. A proportion of coding errors are Kate Pyper PhD https://orcid.org/0000-0002-7782-1048
likely to have occurred in all hospitals but are unlikely to have affected
our conclusions; furthermore, detailed analysis of discharge coding has REFERENCES
suggested these are generally >95% accurate.14 1. Miller TR, Lestina DC. Costs of poisoning in the United States and sav-
We have not considered the effect of National Poisons Informa- ings from poison control centers: a benefit- cost analysis. Ann Emerg
Med. 1997;29(2):239–245.
tion Service’s telephone information services, which are available 24/7
2. Elamin MEMO, James DA, Holmes P, et al. Reductions in emergency
to support clinicians managing more complex poisoning. However, department visits after primary healthcare use of the UK National Poi-
because at least 97% of poison information enquiries from EDs come sons Information Service. Clin Toxicol [Internet]. 2018;56(5):342-347.
via the internet and TOXBASE, differences in telephone call numbers 3. Bateman DN, Good AM, Laing WJ, Kelly CA. TOXBASE: Poisons infor-
mation on the internet. Emerg Med J. 2002;19(1):31-34.
cannot explain the significant differences we have observed. It is also
4. Proudfoot AT, Good AM, Nicholas Bateman D. Clinical toxicology in
likely that most triage decisions occur after review of TOXBASE advice Edinburgh, two centuries of progress. Clin Toxicol. 2013;51(6):509-
and not after a phone call, because these tend to deal with specific 514.
treatment advice.6 5. Bateman DNN, Good AMM. Five years of poisons information
on the internet: The UK experience of TOXBASE. Emerg Med J.
The hospital attendance and admission data did not contain infor-
2006;23(8):614-617.
mation on the agents involved, which may have confounded the 6. UK National Poisons Information Service. NPIS 2017-18 Annual
analysis, particularly for smaller hospitals that receive fewer cases and Report [Internet]. 2017. http://www.npis.org/annualreports.html.
do not tend to handle complex cases poisoning. However, as the data Accessed December 2019.
in England pertain to trusts, differences in case mix should have been 7. Pyper K, Eddleston M, Bateman D, et al. Hospital usage of TOXBASE
in Great Britain: temporal trends in accesses 2008 to 2015. Hum Exp
averaged out. The imputation of small numbers could potentially add
Toxicol. 2018;37:1207-1214.
error, but this is minimized due to the inclusion of many of these smaller 8. Baker C. Accident and Emergency Statistics: Demand, Perfor-
hospitals within the larger Trusts used in our analysis. mance and Pressure [Internet]. 2017. http://researchbriefings.files.
Finally, in the drug group-level analysis, access to specific drug parliament.uk/documents/SN06964/SN06964.pdf. Accessed Decem-
ber 2019.
groups on TOXBASE was used to describe the trend in drugs poison-
9. Bate A, Baker C. NHS Wales Statistics [Internet]. 2016.
ing overall. A better comparison could be made if data were available http://researchbriefings.parliament.uk/ResearchBriefing/Summary/
on admissions and attendances due to poisoning by each specific drug SN06994. Accessed December 2019.
group. However, even if the data were available, the number of admis- 10. Benefit risk profile of acetylcysteine in the management of parac-
etamol overdose [Internet]. Medicines and Healthcare products
sions and attendances would be small, requiring a large amount of
Regulatory Agency. 2012;http://www.mhra.gov.uk/home/groups/pl-
data suppression to maintain patient confidentiality, markedly reduc- p/documents/drugsafetymessage/con184609.pdf
ing usability. 11. Pettie JM, Dow MA, Sandilands EA, Thanacoody HK, Bateman DN. An
integrated care pathway improves the management of paracetamol
poisoning. Emerg Med J. 2012;29(6):482-486.
12. Narayan H, Thomas SHL, Eddleston M, Dear JW, Sandilands E, Bate-
6 CONCLUSIONS man DN. Disproportionate effect on child admissions of the change in
Medicines and Healthcare Products Regulatory Agency guidance for
In summary, we have shown that increased hospital use of an internet- management of paracetamol poisoning: an analysis of hospital admis-
sions for paracetamol overdose in England and Scotland. Br J Clin Phar-
based poisons information database, TOXBASE, is associated with an
macol. 2015;80(6):1458-1463.
increase in overall admission rates overall, but rates of access to par- 13. HSCIC. Statistics on drugs misuse [Internet]. 2016. https://digital.nhs.
ticular drug groups differ in their relationships with admission rates. uk/data-and-information/publications/statistical/statistics-on-drug-
These findings suggest a possible direct effect of TOXBASE on clin- misuse. Accessed December 2019.
ician behavior and patient management decisions and highlight the 14. Burns EM, Rigby E, Mamidanna R, et al. Systematic review of discharge
coding accuracy. J Public Health (Bangkok). 2012;34(1):138-148.
potential use of point-of-care internet poisons information systems.
12 PYPER ET AL .

AUTHOR BIOGRAPHY SUPPORTING INFORMATION

Additional supporting information may be found online in the Support-
Kate Pyper, PhD, is a teaching assistant ing Information section at the end of the article.
in the Mathematics and Statistics depart-
ment at the University of Strathclyde,
Glasgow, Scotland. How to cite this article: Pyper K, Robertson C, Eddleston M,
Sandilands E, Bateman DN. Use of the online poisons
information database TOXBASE and admissions rates for
poisoned patients from emergency departments in England
and Wales during 2008 to 2015. JACEP Open. 2020;1–12.
https://doi.org/10.1002/emp2.12116