OUP user menu

Predicting outcome in acute organophosphorus poisoning with a poison severity score or the Glasgow coma scale

J.O.J. Davies, M. Eddleston, N.A. Buckley
DOI: http://dx.doi.org/10.1093/qjmed/hcn014 371-379 First published online: 4 March 2008


Background: Organophosphorus (OP) pesticide poisoning kills around 200 000 people each year, principally due to self-poisoning in the Asia-Pacific region.

Aim: We wished to assess whether patients at high risk of death could be identified accurately using clinical parameters soon after hospital admission.

Design: We evaluated the usefulness of the International Program on Chemical Safety Poison Severity Score (IPCS PSS) and the Glasgow Coma Score (GCS) prospectively for predicting death in patients poisoned by OP pesticides.

Methods: Data were collected as part of a multicenter cohort study in Sri Lanka. Study doctors saw all patients on admission, collecting data on pulse, blood pressure, pupil size, need for intubation and GCS.

Results: Of the patients, 1365 with a history of acute OP poisoning were included. Receiver operating characteristic (ROC) curves were calculated for the IPCS PSS and GCS on admission. The IPCS PSS and GCS had similar ROC area under the curves (AUC) and best cut points as determined by Youden's index (AUC/sensitivity/specificity 0.81/0.78/0.79 for IPCS PSS ⩾ grade 2 and 0.84/0.79/0.79 for GCS ⩽13). The predictive value varied with the pesticide ingested, being more accurate for dimethoate poisoning and less accurate for fenthion poisoning (GCS AUC 0.91 compared with 0.69).

Conclusions: GCS and the IPCS PSS were similarly effective at predicting outcome. Patients presenting with a GCS ⩽13 need intensive monitoring and treatment. However, the identity of the organophosphate must be taken into account, since the half of all patients who died from fenthion poisoning only had mild symptoms at presentation.


Organophosphorus (OP) pesticide self-poisoning is estimated to kill around 200 000 people each year, largely in the Asia-Pacific region. This predominantly occurs in rural communities and is often an impulsive act comparable to self poisoning with medication in the west; the critical difference being the 10–20% case fatality rate (compared to 0.3% in Britain for example).1 OP poisoning is also of great interest to developed countries vulnerable to terrorist or military attack with nerve agents.2 The principal pharmacological action of all OPs is the inhibition of acetylcholinesterase; most patients die from cardiorespiratory failure. However, there is much variation in the timing of onset and clinical features depending on the particular OP3 involved.

OP poisoning has a high inpatient mortality and many patients have cardiorespiratory arrests after admission (38% of patients requiring intubation in one study4). In this study, we wished to investigate whether it was possible to predict inpatient mortality in OP poisoning using a scoring system based on simple clinical parameters recorded solely at admission. This might enable clinicians to identify patients at high risk of dying soon after presentation, allowing more intensive monitoring and treatment. A simple system based on clinical features is likely to be most useful in low income countries where the majority of OP poisoning occurs.1

A number of systems have been proposed for predicting outcome in OP poisoning. Many are reliant on laboratory tests 5–9 and are, therefore, less useful in resource poor locations. Others that use clinical parameters 10 have only been validated using small numbers of patients.

The International Program on Chemical Safety (IPCS)/EC/EAPCCT Poison Severity Score (IPCS PSS) was developed by the International Program on Chemical Safety, the European Community and the European Association of Poisons Centers and Clinical Toxicologists to create a scoring system that produces a qualitative evaluation of the morbidity caused by different forms of poisoning.11 It has a several different categories, which encompass a large number of clinical features (ranging from the effects of bites and stings to methemaglobin levels) and it is designed to be used flexibly to incorporate the most relevant clinical and laboratory features of the poisoning and data available. This has been evaluated prospectively in one study which found it to be useful in identifying serious and complicated cases of poisoning.12 We wished to see if this score and the Glasgow coma score (GCS) alone could predict mortality in OP pesticide poisoning using data collected prospectively on a patient's admission to a secondary hospital in Sri Lanka.


The prospective data were collected between March 2002 and November 2004 in a cohort study conducted by the South Asian Clinical Toxicology Research Collaboration (SACTRC) in Sri Lanka. The study contains two nested randomized controlled trials (RCTs) of activated charcoal and pralidoxime, and has ethics review committee approval from Oxford and Colombo. Data were collected on patients with poisoning admitted to Anuradhapura, Polonnaruwa and Kurunegala General Hospitals.3,,13 These hospitals receive about 50% of all OP poisoned patients presenting to peripheral hospitals in their districts, and 80% of those who ultimately die.14

All poisoned patients were seen on admission by study doctors. Patients were included in this study if they had a history of OP pesticide ingestion as indicated by the patient or relatives, the transferring doctor or the pesticide bottle. With the exception of concomitant alcohol consumption, patients who had taken more than one poison were excluded. It was often clear from the history on admission that the patient had taken an OP pesticide but the identity of the particular OP taken was sometimes only confirmed later by locating the bottle or relatives. The accuracy of identifying the poison ingested from the history has been validated in a previous study3 in which blood samples were taken from 444 patients on admission. This showed that the history was confirmed by identification of the OP taken in the blood in 95.5% of chlorpyrifos, 85.3% of dimethoate and 95.2% of fenthion patients with BuChE inhibition.

Data were collected prospectively by the study doctors, who recorded the poison ingested and examination findings, including pulse, blood pressure, pupil size and GCS. This examination took place during initial resuscitation with IV atropine, fluids and oxygen. Atropine was often given to symptomatic patients by the transferring hospital.

Some patients were then consented into one or both of two RCTs. The RCT of activated charcoal found no difference in outcome between the treatment arms.15

All patients received standard medical treatment under the direction of the hospitals’ consultant physicians. This followed a standard protocol,16 which was dictated by the patients clinical condition and was independent of the identity of the OP involved. The protocol included rapid atropinization, with doubling doses of atropine at 5–10 min intervals, starting at 1–3 mg, given until muscarinic signs were abolished. The decision to intubate and transfer to the intensive care unit was made by the medical team, independent of the study doctors. All symptomatic patients (apart from those recruited into the pralidoxime trial) received pralidoxime chloride 1 g IV four times a day for 1–3 days.

The data were used to test the value of the IPCS PSS and GCS prospectively. A sample size calculation was performed, which predicted that the width of the 95% CI around estimates of sensitivity of 0.9 and specificity of 0.8 would be less than ±0.05 with a sample size of 1000 patients and a 15% mortality. The IPCS PSS 11 was calculated for each patient using the criteria in Table 1, with the overall score being the patient's most severe grading in any category on admission (missing data were not graded). We did not incorporate oxygen saturation, respiratory rate, ECG findings or more complex investigations such as arterial blood gases into the IPCS PSS; however, the score is designed to incorporate the available data to give an overall impression of the severity of the poisoning.

View this table:
Table 1

Criteria used to calculate the IPCS PSS

Grade 1Grade 2Grade 3
Bradycardia (Pulse)>5041–50⩽40
Tachycardia (Pulse)⩽140141–180>180
Hypotension (Systolic BP)>10081–100⩽80
  • The highest grade scored in any category dictated the overall grade.

A second analysis was performed to investigate whether the particular OP ingested affected the ability to predict outcome. The data were also analyzed to delineate which variables had the greatest predictive value. Sensitivity and specificity were calculated for the GCS, pulse, systolic blood pressure, pupil size and need for intubation. Receiver operating characteristic (ROC) curves for predicting inpatient mortality were generated using this data. The sensitivities, specificities, likelihood ratios, odds ratios and area under the curve (AUC), an indicator of the predictive value of a test, were calculated using GraphPad Prism V4.0. The best cut-off values were calculated as the point having the maximal Youden's index (Youden's index = sensitivity + specificity − 1).17


The primary analyses looked at 1365 patients who presented to hospital with a diagnosis of probable OP poisoning and then 990 patients in whom the poison was subsequently specifically identified as one of chlorpyrifos, fenthion, or dimethoate. The median time between ingestion and presentation was 4 h (interquartile range 3–5 h).

The IPCS PSS on admission was useful in predicting outcome in patients identified as having taken OPs on admission, with grade 3 having a sensitivity of 0.66 and specificity of 0.88 for predicting death and a score of grade 2 or more having sensitivity and specificity of 0.78 and 0.79, respectively.

The GCS was as good a predictor of outcome as the IPCS PSS itself (AUC 0.84 vs 0.81) and level of consciousness made the greatest contribution to the IPCS PSS (Table 3, Figure 1). A GCS of ⩽13 was the best predictor overall with a sensitivity and specificity both of 0.79 (95% CI: 0.72–0.84 and 0.77–0.82, respectively).

Figure 1.

ROC plot for all OPs comparing the predictive value of GCS, pulse, blood pressure, pupil size and intubation.

Further analysis revealed marked differences between the different OPs (Tables 2 and 3, Figures 2–4). Mortality could be most accurately predicted on admission in patients with dimethoate poisoning and was difficult to predict in fenthion poisoning, with half of all fatal fenthion cases presenting with a grade 1 IPCS PSS score.

Figure 2.

ROC plot of the IPCS PSS's variable ability to predict death for different OPs.

Figure 3.

ROC plot comparing the ability of GCS to predict outcome for different OPs.

Figure 4.

Plot of the odds ratios for the best cut offs for the IPCS PSS, GCS, systolic BP, pulse and intubation in the different OPs.

View this table:
Table 2

Comparison of the IPCS PSS and overall optimal GCS cut off

All OP n (deaths of total)Fenthion n (deaths of total)Chlorpyrifos n (deaths of total)Dimethoate n (deaths of total)
IPCS PSSGrade 141 of 973 (4%)8 of 100 (8%)16 of 417 (4%)4 of 186 (2%)
Grade 222 of 138 (6%)4 of 12 (33%)6 of 55 (11%)4 of 35 (11%)
Grade 3121 of 254 (47%)4 of 11 (36%)21 of 61 (34%)71 of 113 (63%)
GCS>1339 of 974 (4%)9 of 102 (8%)13 of 407 (3%)5 of 182 (3%)
⩽13145 of 390 (37%)7 of 21 (33%)30 of 126 (24%)74 of 147 (50%)
Overall184 of 1365 (13%)16 of 123 (13%)43 of 533 (8%)79 of 334 (24%)
View this table:
Table 3

Comparison of best cut-off values for IPCS PSS, GCS, pulse, blood pressure and intubation

All OPsFenthionChlorpyrifosDimethoate
IPCS PSS cut off Grade 2 or 3
AUC0.81 (0.77–0.85)0.67 (0.52–0.84)0.75 (0.65–0.83)0.88 (0.84–0.92)
Sensitivity0.78 (0.71–0.84)0.50 (0.25–0.75)0.63 (0.47–0.77)0.95 (0.99–0.99)
Specificity0.79 (0.76–0.81)0.86 (0.78–0.92)0.82 (0.78–0.85)0.71 (0.65–0.77)
Likelihood ratio3.
n (% missing data)1365 (0)123 (0)533 (0)334 (0)
GCS cut off ⩽13
AUC0.84 (0.80–0.87)0.69 (0.54–0.85)0.78 (0.71–0.87)0.91 (0.87–0.95)
Sensitivity0.79 (0.72–0.84)0.43 (0.20–0.70)0.70 (0.54–0.83)0.94 (0.79–0.95)
Specificity0.79 (0.77–0.82)0.87 (0.79–0.93)0.80 (0.77–0.84)0.71 (0.79–0.88)
Likelihood ratio3.
n (% missing data)1364 (0.1)123 (0)533 (0)334 (0)
Systolic BP cut off ⩽100
AUC0.64 (0.59–0.70)0.41 (0.34–0.58)0.52 (0.40–0.64)0.81 (0.74–0.87)
Sensitivity0.45 (0.37–0.52)0.19 (0.04–0.46)0.24 (0.11–0.40)0.69 (0.57–0.79)
Specificity0.85 (0.84–0.88)0.84 (0.75–0.90)0.88 (0.84–0.91)0.82 (0.76–0.87)
Likelihood ratio3.
n (% missing data)1259 (7.8)115 (6.5)494 (7.3)311 (6.8)
Pulse cut off ⩽60
AUC0.44 (0.40–0.50)0.40 (0.33–0.58)0.49 (0.38–0.59)0.45 (0.38–0.53)
Sensitivity0.09 (0.05–0.15)0.0 (0.0–0.22)0.18 (0.08–0.34)0.09 (0.04–0.18)
Specificity0.97 (0.53–0.59)0.96 (0.91–0.99)0.98 (0.96–0.99)0.98 (0.96–0.99)
Likelihood ratio3.
n (% missing data)1328 (2.7)121 (1.63)524 (1.7)324 (3.0)
AUC0.66 (0.61–0.71)0.53 (0.37–0.69)0.60 (0.50–0.70)0.72 (0.65–0.79)
Sensitivity0.38 (0.30–0.45)0.06 (0.00–0.30)0.23 (0.12–0.39)0.49 (0.38–0.61)
Specificity0.95 (0.94–0.96)1.00 (0.97–1.00)0.97 (0.95–0.98)0.95 (0.91–0.97)
Likelihood ratio7.96.79.6
n (% missing data)1365 (0)123 (0)533 (0)334 (0)
  • The range in parenthesis denotes the 95% CI of the sensitivity, specificity and AUC of the ROC.

Using the cut off of GCS ⩽13, 39 of the 184 deaths (21%) would not have been predicted on admission. This varied markedly between the different OPs with only 5 of 79 deaths (6%) being missed in dimethoate poisoning compared with 13 of 43 (30%) in chlorpyrifos and 9 of 16 deaths (56%) in fenthion poisoning. The predictive value of the GCS was improved if different cut offs were used for the different poisons. This was particularly marked in dimethoate poisoning where reducing the cut off to GCS ⩽10 increased the likelihood ratio from 3.3 to 5.0 [sensitivity 0.92 (95% CI: 0.84–0.97), specificity 0.82 (95% CI: 0.76–0.86)].

Although the decision to intubate on admission was carried out by the local doctors without defined criteria it had a specificity of 95% for predicting mortality; however, it had a low sensitivity, particularly in fenthion poisoning (sensitivity 0.06, 95% CI: 0.00–0.30).

A systolic blood pressure ⩽100 mmHg in dimethoate-poisoned patients had a sensitivity of 0.69 (95% CI: 0.32–0.55) and specificity of 0.82 (95% CI: 0.94–0.99). This effect was not seen with the other OPs; the AUC was 0.81 for dimethoate, but only 0.52 and 0.41 for chlorpyrifos and fenthion, respectively.

The heart rate had little value in predicting outcome, with an AUC of 0.44. There was an excess of mortality in a small number patients with marked bradycardia (Figure 5). This was statistically significant, but the low sensitivity meant that it did not contribute greatly to the predictive value of the IPCS PSS or GCS (of the 16 patients with a pulse ⩽60 only one had a GCS >13). The high likelihood and odds ratios are due to the significant specificity and give a false impression of the usefulness of this parameter.

Figure 5.

Histogram of the pulse on admission in survivors and deaths.

Pupil size on admission was also of little value with AUC of 0.51 (Figure 1). No patients in our series had seizures on admission.


This study shows that the GCS and IPCS PSS on admission are able to predict death in patients with presumed OP poisoning with similar accuracy, however, the GCS is much easier to apply clinically. Using a cut off for GCS ⩽13 the patients could be divided into a high risk group with a 37% case fatality rate and a low risk group with only a 4% mortality. The accuracy of both scores varied markedly depending on the poison taken, being highly discriminative in dimethoate poisoning and less accurate in fenthion poisoning, where around half of all patients who went on to die were identified as being low risk at presentation.

In dimethoate poisoning the patients presenting with GCS 14 or more had a mortality of 2.7% compared to a case fatality rate of 50.3% if the GCS was ⩽ 13 at presentation. We believe this effect was principally due to the delay in onset of symptoms after ingestion. In dimethoate poisoning, severely poisoned patients are usually symptomatic by the time they reach medical facilities and the GCS is both sensitive and specific at predicting outcome.

Both the GCS and the IPCS PSS score had a poor sensitivity in fenthion poisoning. This was probably because the onset of symptoms in fenthion poisoning is often delayed for up to several days after admission to hospital18 and the predominant toxicity is delayed onset respiratory paralysis.3 It is unlikely that any system based on clinical criteria at presentation would be highly sensitive. However, early onset of symptoms in fenthion poisoning is an indicator of severe toxicity; with patients presenting with a reduced GCS or high IPCS PSS having a greater mortality rate than in chlorpyrifos poisoning (Table 2).

This study used data collected on admission and did not evaluate the ability of the scoring systems to predict outcome after admission. It is likely that the specificity for predicting outcome would be retained if they were used in this way but their sensitivity may fall, since patients who develop delayed onset respiratory paralysis could be missed.

Previous scoring systems are summarized in Table 4. Most use complex laboratory investigations that are not available in many of the settings where OP poisoning occurs. Two studies have compared GCS with APACHE II and Simplified acute physiology score (SAPS) II.19,,20 These have shown no significant differences between the AUC or optimal sensitivities and specificities for GCS compared with the more complex scoring systems; however, both studies had very small sample sizes, and did not have the statistical power to detect even 3-fold differences in the performance of these scores (Table 4).

View this table:
Table 4

Previous scoring systems in OP poisoning

ReferenceParameters used in grading systemValidationn (deaths)Ability to predict mortality
Sensitivity (95% CI)Specificity (95% CI)
Bardin et al.5Clinical findings (rhonchi, consciousness, hypotension), CXR, ABGRetrospective62 (3)1.00* (0.29–1.00)0.47* (0.34–0.60)
Bilgin et al.19APACHE II >16Retrospective32 (7)1.00 (0.59–1.00*)0.80 (0.59–0.93*) AUC 0.92
SAPS II >29Retrospective32 (7)1.00 (0.59–1.00*)0.76 (0.55–0.91*) AUC 0.89
GCS <9Retrospective32 (7)0.86 (0.42–1.00*)0.72 (0.51–0.88*) AUC 0.90
Goel et al.6Clinical findings (GCS, pupil size, fasciculations, respiratory failure)Retrospective103 (9)1.00* (0.66–1.0)0.65* (0.54–0.74)
Grmec et al.7GCS ⩽ 6Retrospective65 (4)0.84 (0.76–0.91)0.89 (0.84–0.93)
QTc on ECGRetrospective65 (4)0.90 (0.84–0.93)0.83 (0.71–0.93)
Lee and Tai8APACHE II >26Retrospective23 (3)1.00* (0.29–1.00)0.95* (0.75–1.00)
Sennayake et al.10Clinical findings (Miosis, fasciculations, respiratory rate, bradycardia, level of consciousness)Prospective173 (10)0.80* (0.44–0.97)0.86* (0.80–0.91)
Sungurtekin et al.20APACHE II >11Retrospective48 (11)0.90 (0.56–0.98)0.60 (0.42–0.75) AUC 0.79
SAPS II >28Retrospective48 (11)0.90 (0.56–0.98)0.81 (0.65–0.92) AUC 0.85
GCS <9Retrospective48 (11)0.64 (0.31–0.89)0.78 (0.62–0.91) AUC 0.64
  • *Values for sensitivity (95% CI), specificity (95% CI) and AUC calculated from the published data.

Several studies6,,10 have used criteria such as miosis and bradycardia, which we found to be of little value in predicting outcome in OP poisoning. Some9 have been based on indicators of poor outcome that include treatment choices, such as the need for intubation or dosage of atropine (without standardized criteria for these interventions). This creates circular arguments that cannot be applied prospectively.

A limitation of this study is that we did not incorporate respiratory rate or pulse oximetry into our analysis (unfortunately it was not recorded with enough reliability). The use of respiratory parameters to predict outcome in OP poisoning is likely to be difficult. The scale proposed by Senanayake, for example, uses high respiratory rate (>20) and the presence of cyanosis. This approach is may cause difficulties since severe OP poisoning can cause either central respiratory depression with a reduced respiratory rate or tachypneoa in the context of bronchorrhea, bronchoconstriction or respiratory muscle weakness.4 Studies using cyanosis,10 pulse oximetry or ABG5 readings, without measuring the inspired oxygen concentration are likely to be flawed since altering the inspired oxygen changes these parameters markedly. The alveolar arterial oxygen gradient as is used by the APACHE II21 scoring system circumvents this problem. This has only been used in retrospective studies.8,,19,20 Controlled oxygen delivery and arterial blood gas measurements are not available in hospitals where most OP-poisoned patients present and a scoring system that requires the calculation of the alveolar arterial oxygen gradient becomes complex and difficult to use in clinical practice.

Intubation on admission was highly specific at predicting mortality but not very sensitive. It also did not add to the predictive value of the GCS, because nearly all patients who were intubated also had a reduced GCS.

The utility of heart rate and pupil size as predictors may have been reduced by the frequent use of atropine prior to assessment. It is essential to administer atropine early to patients16,,22 and this should be done at the peripheral clinic or during transfer to hospital. Any practical scoring system, therefore, needs to be robust in patients treated and untreated with atropine.

The differing predictive utility of hypotension between the OPs reflects our clinical experience that dimethoate can cause an almost universally fatal syndrome characterized by progressive hypotension that only partially responds to IV fluids and inotropes.3 While this syndrome is unusual in chlorpyrifos and fenthion poisoning, hypotension unresponsive to atropine and IV fluids is likely to be an ominous feature in any form of OP poisoning.

Our results show that the ability to predict outcome at presentation varies with the specific OP. This raises problems with the applicability of any scoring system not designed to take the identity of the OP into account. It is unrealistic to design scoring systems that incorporate the differing effects of each and every individual OP, since there are many OPs and little data for individual OPs. However, we may be able to make some extrapolations. It is difficult to predict outcome in fenthion poisoning because it has delayed effects and this is likely to be true for other highly lipid soluble OPs with a slow onset of action. Patients with fenthion poisoning have required intubation up to 113 h after admission to hospital4 and we would recommend that clinicians consider monitoring symptomatic patients for 5 days. Asymptomatic patients can probably be discharged after 2–3 days. It was easier to predict outcome in dimethoate poisoning and it is probable that this would also apply to other OPs with a very small volume of distribution or rapid onset of action. It is important to remember that even in dimethoate-poisoning patients occasionally need to be intubated up to 100 h after admission4 and in this study 2% of patients who went on to die had normal severity scores at presentation. It is probably safe to discharge asymptomatic patients with dimethoate poisoning after 1–2 days.

These results may be of relevance to OP nerve gases. Terrorist or military attack with nerve agents is likely to stretch even well resourced health services and a clinical system for triage should be useful. After the sarin attacks on the Tokyo subway, in which 12 people died and less than 1000 people were injured,23 more than 4000 patients presented to medical facilities. The majority were essentially asymptomatic.24

Nerve agents are likely to have a more rapid onset and possibly shorter duration of action than OP pesticides,25 although delayed onset of symptoms is described after dermal contamination and there are reports of a small number of late deaths.26 It is probable that patients with a reduced GCS or high IPCS PSS on admission would have a high mortality in nerve agent poisoning and that asymptomatic patients at presentation would probably survive, providing that they are decontaminated.

ROC analyses of each of the components of the modified IPCS PSS allowed the diagnostic contribution of each component of the score to be compared. However, only the GCS has been suggested to be a predictor of outcome when used on its own. Therefore, the usual thresholds for ROC curves (e.g. AUC >0.75 for a test to be at all useful) may be too stringent for each component of the score. However, it is clear that the pulse adds little to the prognostic score and the GCS is clearly the most informative component of the assessment from a prognostic perspective.


In our study, the GCS was as good as the IPCS PSS. Patients who present with a GCS ⩽13 need intensive monitoring and treatment. It is crucial that the identity of the OP be taken into account because highly lipid soluble poisons such as fenthion can cause delayed effects and half of all patients that die from this type of poisoning only have mild symptoms at presentation. Patients poisoned with such OP, therefore, need close monitoring even if they are asymptomatic at presentation.


We thank the members of the Ox-Col collaboration for their hard work collecting the data, and also the directors, physicians, medical and nursing staff of the study hospitals for their help and enthusiasm. We thank Prof D. Gunnell for statistical review. This work was funded by grant GR063560MA from the Wellcome's Tropical Interest Group to ME, who is a Wellcome Trust Career Development Fellow. SACTRC is funded by the Wellcome Trust/National Health and Medical Research Council International Collaborative Research Grant GR071669MA.

Conflict of interest: None declared.

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (<http://creativecommons.org/licenses/by-nc/2.0/uk/>) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.


View Abstract