European Journal of Cardiovascular Medicine

Aluminium phosphide (AlP) is a cheap and effective solid fumigant used to protect stored grain. It is sold in India as 3 g tablets under trade names such as Celphos and Quickphos. Since the first large clinical series from northern India in the 1980s, AlP has become one of the commonest agents of suicidal poisoning in the region [1]. Hospital data from Haryana showed a steady rise in admissions, with a mortality of more than 75% [2]. A 25-year autopsy study from Chandigarh found that AlP had replaced other agents as the leading cause of fatal poisoning [3]. Pesticide ingestion as a whole accounts for a large share of suicides in developing countries [4], and AlP poisoning is also a major problem in Iran [5].

On contact with water and gastric acid, AlP releases phosphine gas. Phosphine is a cellular poison that impairs mitochondrial respiration and causes widespread oxidative injury [6]. Inhibition of cytochrome c oxidase has been shown in the platelets of poisoned patients [7]. The heart is the main target organ. Refractory shock, arrhythmias and ECG changes of toxic myocarditis are common in severe cases [8]. There is no specific antidote, and management depends on early decontamination, correction of acidosis and haemodynamic support in an intensive care unit [9,10].

Because treatment is mainly supportive and intensive care beds are limited, it is important to recognise early the patients most likely to die. A previous systematic review described the toxicology, clinical features and management of AlP poisoning, but it did not pool prognostic data [11]. Individual studies have reported many possible predictors, including shock, metabolic acidosis, altered consciousness, ECG abnormalities, dose and severity scores [12,13]. These studies were small, and their results were not always consistent. We therefore carried out a systematic review and meta-analysis of the clinical and laboratory predictors of mortality in acute AlP poisoning.

Protocol and reporting

The review followed the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines [14] and is reported according to the PRISMA statement [15]. The protocol was not registered in advance. Ethics committee approval was not required, as the review used only published data.

Search strategy

PubMed/MEDLINE was searched from inception to 31 December 2015 using the following strategy: ("aluminium phosphide" OR "aluminum phosphide" OR celphos OR "rice tablet") AND (mortality OR survivors OR prognosis OR prognostic OR outcome OR predictor OR "non-survivors"). A full-text search of PubMed Central was run with similar terms to find studies in which outcome comparisons appeared only in the body of the paper. The reference lists of included studies and of relevant reviews [6,9,10,11] were checked by hand. No language restriction was applied.

Eligibility criteria

Studies were included if they (1) enrolled patients with acute AlP poisoning confirmed by history and clinical features; (2) reported in-hospital mortality; and (3) compared at least one clinical or laboratory variable between survivors and non-survivors, or reported its association with death. Prospective and retrospective cohort, cross-sectional and case-control studies were eligible. Case reports, case series of fewer than ten patients, animal and laboratory studies, reviews, autopsy-only series and studies of mixed poisons without separate AlP data were excluded.

Study selection and data extraction

Titles and abstracts were screened first, and the full texts of potentially eligible records were then assessed. Data were extracted on a pre-designed form and re-checked against the source article. Extracted items were first author, year, country, design, sample size, number of deaths and every predictor reported for survivors and non-survivors. For continuous variables, the mean, standard deviation and number in each group were taken. For dichotomous variables, the number of patients with the factor in each group was taken. Where a study reported values at several time points, the admission value was used.

Quality assessment

The methodological quality of each study was assessed with the Newcastle-Ottawa Scale (NOS) adapted for cohort studies [16]. The scale awards up to four points for selection, two for comparability and three for outcome, giving a maximum of nine. Studies scoring 7 to 9 were taken as good quality, 5 to 6 as moderate and 4 or less as poor.

Statistical analysis

A predictor was pooled when at least two studies reported comparable data. Continuous variables measured on the same scale (pH, bicarbonate and time to hospital) were pooled as mean differences (MD) between non-survivors and survivors. Variables measured in different units or on different scales (ingested dose in tablets or grams, and severity scores) were pooled as standardised mean differences (SMD, Hedges' g). Dichotomous variables were pooled as odds ratios (OR) for death. A continuity correction of 0.5 was added to all cells of studies with a zero cell. Because of expected clinical and methodological differences, all analyses used the DerSimonian and Laird random-effects model [17]. Heterogeneity was assessed with the chi-square test and quantified with the I² statistic; values of 25%, 50% and 75% were taken as low, moderate and high [18]. Sensitivity analyses were done by omitting one study at a time and by excluding the paediatric study. Funnel plots and Egger's test were planned, but they were not done because no analysis included ten or more studies [19,20]. Analyses were performed in Review Manager (RevMan) version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2014). A p value below 0.05 was taken as significant.

Study selection

The searches identified 137 records, of which 124 remained after duplicates were removed (Figure 1). After title and abstract screening, 59 full-text articles were assessed. Of these, 40 were excluded: 10 were mixed-poison series without separate AlP outcome data, 6 were treatment trials without a prognostic comparison, 10 were biochemical or mechanistic studies, 5 were fatal or autopsy series only and 9 were reviews or descriptive series. Nineteen studies were included in the qualitative synthesis [2,5,12,13,21-35], and six of them provided data suitable for meta-analysis [13,30,31,33,34,35].

AlP: aluminium phosphide.

Study characteristics and quality

The 19 studies were published between 1986 and 2014 and together described about 1,826 patients (Table 1). Twelve were from India, five from Iran and two from Morocco. Ten were prospective, eight were retrospective and one did not state its design. One study was limited to children [34]. Reported mortality ranged from 18.6% to 77.2%. On the Newcastle-Ottawa Scale, two studies were of good quality, fifteen of moderate quality and two of poor quality. The main weaknesses were single-centre samples, small numbers of patients and lack of adjustment for confounders. Only one study used multivariate analysis [12].

Table 1: Characteristics and quality of the 19 included studies

Included in the meta-analysis. * Patients with severe or mild toxicity. NR: not reported in the source; NOS: Newcastle-Ottawa Scale (maximum 9); ARDS: acute respiratory distress syndrome; DIC: disseminated intravascular coagulation; MV: mechanical ventilation; GCS: Glasgow coma scale; SAPS: Simplified Acute Physiology Score; APACHE: Acute Physiology and Chronic Health Evaluation; PRISM: Pediatric Risk of Mortality.

Metabolic acidosis

Four studies with 215 patients reported admission pH [13,30,33,35]. Non-survivors had a significantly lower pH than survivors (MD −0.13, 95% CI −0.17 to −0.08, p < 0.001), with no heterogeneity (I² = 0%) (Figure 2). Two studies with 73 patients reported serum bicarbonate [13,35]. Bicarbonate was 5.28 mmol/L lower in non-survivors (95% CI −7.99 to −2.57, p < 0.001; I² = 0%) (Figure 3). These results agree with a large Tehran series, in which all patients with pH below 7.0 died and all with pH of 7.35 or more survived [5].

Mean difference (non-survivors minus survivors), inverse-variance random-effects model.

Mean difference (non-survivors minus survivors), inverse-variance random-effects model.

Shock and haemodynamic failure

Three studies with 97 patients reported shock or hypotension at presentation [13,30,34]. Shock was present in 46 of 48 non-survivors (95.8%) and 25 of 49 survivors (51.0%). It was associated with higher odds of death (OR 11.87, 95% CI 1.12 to 126.38, p = 0.040), with moderate heterogeneity (I² = 59%) (Figure 4). The wide confidence interval reflects the small number of survivors without shock. In the narrative synthesis, shock was the strongest independent predictor of death in the only multivariate analysis (RR 3.82, 95% CI 1.12 to 13.38) [12]. It was also a severity factor in the earlier Moroccan series [28]. Poor response to dopamine [2], an abnormal heart rate response in shock [22] and the need for vasoactive drugs [12,13,28] were further markers of fatal outcome.

Odds ratio, inverse-variance random-effects model; a continuity correction of 0.5 was applied to Jaiswal 2009, in which shock was taken as blood pressure not recordable at admission. Values in the n columns are events/total.

Ingested dose and severity scores

Four studies with 192 patients reported the amount of AlP ingested [13,30,33,34]. Non-survivors had taken a larger dose (SMD 0.79, 95% CI 0.07 to 1.50, p = 0.032), but heterogeneity was high (I² = 76%) (Figure 5). One study found no difference in dose [13]. Three studies with 96 patients reported a severity score at admission: SAPS II [31], APACHE II [13] and PRISM III in children [34]. Scores were markedly higher in non-survivors (SMD 1.70, 95% CI 0.92 to 2.47, p < 0.001; I² = 60%) (Figure 6). Ingestion of fresh rather than exposed tablets was linked with death in three studies [21,26,30]. In the largest of these, all seven patients who took exposed tablets survived [30].

Standardised mean difference (Hedges' g), inverse-variance random-effects model. Dose was reported as number of tablets in Jaiswal 2009 (fresh tablets) and Hosseinian 2011, and in grams in Mathai 2010 and Sharma 2014.

Standardised mean difference (Hedges' g), inverse-variance random-effects model. Scores: SAPS II (Shadnia 2010), APACHE II (Mathai 2010) and PRISM III (Sharma 2014).

Other predictors

Time from ingestion to hospital did not differ significantly between the groups (MD 0.78 h, 95% CI −0.05 to 1.62, p = 0.065; 2 studies, 57 patients). Male sex was not associated with death (OR 0.73, 95% CI 0.30 to 1.76, p = 0.48; 3 studies, 97 patients; I² = 0%). The pooled estimates are summarised in Table 2. Among predictors that could not be pooled, ECG abnormalities were linked with death in two Moroccan studies [12,28] and a large Iranian series [5]. They were also linked with death when they occurred together with hypomagnesaemia [27]. An older Indian series found no effect of ECG changes on mortality [23]. Altered consciousness [12], hyperglycaemia [29], raised methaemoglobin [32], high blood phosphine levels [24], acute renal failure [12] and raised serum creatinine [13] were each reported as markers of poor outcome in single studies. Severe hypotension and metabolic acidosis were also more marked in non-survivors in a large clinico-pathological series [25].

Table 2: Summary of pooled estimates for predictors of mortality (random-effects model)

MD: mean difference (non-survivors minus survivors); SMD: standardised mean difference; OR: odds ratio for death.

Sensitivity analysis

The association of low pH with death was stable when each study was omitted in turn, with pooled MD values from −0.11 to −0.14 and I² of 0% in every case (Table 3). When the paediatric study was excluded, the severity score remained significant (SMD 1.48, 95% CI 0.41 to 2.55), but the dose association lost significance (SMD 0.52, 95% CI −0.17 to 1.20). This suggests that the dose effect was partly driven by children, who ingest more poison relative to body weight. Excluding the study with a zero cell left the direction of the shock association unchanged but made it non-significant (OR 4.18, 95% CI 0.75 to 23.45).

Table 3: Sensitivity analyses

pH values are mean differences. Adult-only analyses exclude Sharma 2014.

This systematic review brings together three decades of clinical research on prognosis in acute AlP poisoning. Low arterial pH and low serum bicarbonate at admission were the most consistent predictors of death, with no heterogeneity between studies. Shock at presentation, a larger ingested dose and a high admission severity score were also associated with death. Sex and time to hospital were not.

The strength and consistency of the acid-base findings fit with the known mechanism of phosphine toxicity. Phosphine blocks mitochondrial oxidative phosphorylation and shifts cells to anaerobic metabolism, which produces lactic acidosis [6,7]. The degree of acidosis therefore acts as a direct measure of the cellular energy failure that drives multi-organ dysfunction. Admission lactate was also higher in non-survivors in one included study [35]. In the Tehran series, pH separated those who lived from those who died almost completely at its extremes [5]. Arterial blood gas analysis is quick and widely available, which makes pH and bicarbonate practical triage tools in district and teaching hospitals.

Shock is the cardinal feature of severe AlP poisoning and reflects both direct myocardial injury and peripheral vascular failure [6,8]. Our pooled odds ratio was large but imprecise, because almost all non-survivors and about half of survivors were in shock. The only multivariate analysis also identified shock as an independent predictor [12]. Refractory hypotension, failure to respond to dopamine and the need for vasoactive drugs were repeatedly reported in fatal cases [2,12,13,28]. Together these findings support early invasive haemodynamic monitoring and intensive care for every patient who presents with hypotension [9].

The amount of poison ingested and its freshness determine the amount of phosphine released. Fresh tablets from a newly opened container were consistently more lethal than exposed tablets [21,26,30]. Ingested dose showed high heterogeneity in our analysis, and its effect was not significant in adults alone. A likely reason is that the reported dose depends on patient or relative recall and does not account for vomiting or tablet freshness. Severity scores such as SAPS II, APACHE II and PRISM III combine several physiological variables and performed well [13,31,34]. These scores, however, take time to calculate and may not be available in emergency departments in resource-limited settings.

Several other markers were reported in single studies, including ECG abnormalities, altered consciousness, hyperglycaemia, methaemoglobinaemia and blood phosphine levels [12,24,29,32]. The role of ECG changes was not consistent. Earlier Indian data showed no effect on mortality [23], but later studies linked them with death, especially when they occurred with hypomagnesaemia [5,12,27,28]. More studies with uniform definitions are needed before these markers can be recommended for routine prognosis.

These findings matter to forensic medicine as well as to clinical care. AlP remains a leading cause of poisoning deaths in medicolegal autopsy practice in India [3]. Recording admission pH, bicarbonate, blood pressure and the form and number of tablets ingested helps the treating team. It also gives a clear record for the forensic expert when the cause and time course of death must be explained. Restricting the sale of AlP tablets and promoting safer formulations remain the most effective preventive measures [4,11].

Strengths and limitations

This is, to our knowledge, the first meta-analysis of predictors of mortality in AlP poisoning. It followed MOOSE and PRISMA guidance, assessed study quality, used random-effects models and tested the stability of results with sensitivity analyses. The review has several limitations. The included studies were small, mostly single-centre and mostly of moderate quality. Only six studies reported data in a form suitable for pooling, so each meta-analysis included only two to four studies. As a result, some estimates were imprecise and publication bias could not be formally tested. Definitions of shock, dose and ECG abnormality varied between studies. Most studies reported only unadjusted comparisons, so the pooled estimates cannot show which predictors are independent of each other. Two Indian series from the same centre may have included overlapping patients [2,23]. The search was limited to PubMed/MEDLINE and PubMed Central, and the review was done by a single author, which may have led to some studies or data being missed.

In acute aluminium phosphide poisoning, a low arterial pH, low serum bicarbonate, shock at presentation, a larger ingested dose and a high severity score at admission predict death. Arterial blood gas analysis and blood pressure measurement at admission are simple, inexpensive tools for identifying high-risk patients who need urgent intensive care. Large prospective multicentre studies with standard definitions and multivariate analysis are needed to build a validated prognostic score for this poisoning.

1. Singh S, Dilawari JB, Vashist R, Malhotra HS, Sharma BK. Aluminium phosphide ingestion. Br Med J (Clin Res Ed). 1985;290(6475):1110-1. PMID: 3921126.
2. Chugh SN, Dushyant, Ram S, Arora B, Malhotra KC. Incidence & outcome of aluminium phosphide poisoning in a hospital study. Indian J Med Res. 1991;94:232-5. PMID: 1937606.
3. Singh D, Jit I, Tyagi S. Changing trends in acute poisoning in Chandigarh zone: a 25-year autopsy experience from a tertiary care hospital in northern India. Am J Forensic Med Pathol. 1999;20(2):203-10. PMID: 10414665.
4. Gunnell D, Eddleston M. Suicide by intentional ingestion of pesticides: a continuing tragedy in developing countries. Int J Epidemiol. 2003;32(6):902-9. PMID: 14681240.
5. Shadnia S, Sasanian G, Allami P, Hosseini A, Ranjbar A, Amini-Shirazi N, et al. A retrospective 7-years study of aluminum phosphide poisoning in Tehran: opportunities for prevention. Hum Exp Toxicol. 2009;28(4):209-13. PMID: 19734272.
6. Proudfoot AT. Aluminium and zinc phosphide poisoning. Clin Toxicol (Phila). 2009;47(2):89-100. PMID: 19280425.
7. Singh S, Bhalla A, Verma SK, Kaur A, Gill K. Cytochrome-c oxidase inhibition in 26 aluminum phosphide poisoned patients. Clin Toxicol (Phila). 2006;44(2):155-8. PMID: 16615671.
8. Soltaninejad K, Beyranvand MR, Momenzadeh SA, Shadnia S. Electrocardiographic findings and cardiac manifestations in acute aluminum phosphide poisoning. J Forensic Leg Med. 2012;19(5):291-3. PMID: 22687771.
9. Gurjar M, Baronia AK, Azim A, Sharma K. Managing aluminum phosphide poisonings. J Emerg Trauma Shock. 2011;4(3):378-84. PMID: 21887030.
10. Moghadamnia AA. An update on toxicology of aluminum phosphide. Daru. 2012;20(1):25. PMID: 23351193.
11. Mehrpour O, Jafarzadeh M, Abdollahi M. A systematic review of aluminium phosphide poisoning. Arh Hig Rada Toksikol. 2012;63(1):61-73. PMID: 22450207.
12. Louriz M, Dendane T, Abidi K, Madani N, Abouqal R, Zeggwagh AA. Prognostic factors of acute aluminum phosphide poisoning. Indian J Med Sci. 2009;63(6):227-34. PMID: 19602756.
13. Mathai A, Bhanu MS. Acute aluminium phosphide poisoning: Can we predict mortality? Indian J Anaesth. 2010;54(4):302-7. PMID: 20882171.
14. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA. 2000;283(15):2008-12. PMID: 10789670.
15. Moher D, Liberati A, Tetzlaff J, Altman DG, PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 2009;6(7):e1000097. PMID: 19621072.
16. Stang A. Critical evaluation of the Newcastle-Ottawa scale for the assessment of the quality of nonrandomized studies in meta-analyses. Eur J Epidemiol. 2010;25(9):603-5. PMID: 20652370.
17. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7(3):177-88. PMID: 3802833.
18. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-60. PMID: 12958120.
19. Sterne JA, Sutton AJ, Ioannidis JP, Terrin N, Jones DR, Lau J, et al. Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials. BMJ. 2011;343:d4002. PMID: 21784880.
20. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315(7109):629-34. PMID: 9310563.
21. Chopra JS, Kalra OP, Malik VS, Sharma R, Chandna A. Aluminium phosphide poisoning: a prospective study of 16 cases in one year. Postgrad Med J. 1986;62(734):1113-5. PMID: 3658848.
22. Chugh SN, Ram S, Singhal HR, Malhotra KC. Significance of heart rate response in shock due to aluminium phosphide poisoning. J Assoc Physicians India. 1989;37(11):708-9. PMID: 2632538.
23. Chugh SN, Chugh K, Ram S, Malhotra KC. Electrocardiographic abnormalities in aluminium phosphide poisoning with special reference to its incidence, pathogenesis, mortality and histopathology. J Indian Med Assoc. 1991;89(2):32-5. PMID: 2056173.
24. Chugh SN, Pal R, Singh V, Seth S. Serial blood phosphine levels in acute aluminium phosphide poisoning. J Assoc Physicians India. 1996;44(3):184-5. PMID: 9251315.
25. Singh S, Singh D, Wig N, Jit I, Sharma BK. Aluminum phosphide ingestion--a clinico-pathologic study. J Toxicol Clin Toxicol. 1996;34(6):703-6. PMID: 8941200.
26. Mohanchand R, Hasija PK, Pradhan AB, Pahwa RS. Prognostic factors in aluminium phosphide poisoning. Med J Armed Forces India. 1997;53(4):274-6. PMID: 28769511.
27. Singh GP. Mortality in acute aluminium phosphide poisoning having hypomagnesaemia with and without ECG changes. J Indian Med Assoc. 2000;98(8):446. PMID: 11294325.
28. Hajouji Idrissi M, Oualili L, Abidi K, Abouqal R, Kerkeb O, Zeggwagh AA. Severity factors of aluminium phosphide poisoning (Phostoxin) [Article in French]. Ann Fr Anesth Reanim. 2006;25(4):382-5. PMID: 16488104.
29. Mehrpour O, Alfred S, Shadnia S, Keyler DE, Soltaninejad K, Chalaki N, et al. Hyperglycemia in acute aluminum phosphide poisoning as a potential prognostic factor. Hum Exp Toxicol. 2008;27(7):591-5. PMID: 18829736.
30. Jaiswal S, Verma RK, Tewari N. Aluminum phosphide poisoning: effect of correction of severe metabolic acidosis on patient outcome. Indian J Crit Care Med. 2009;13(1):21-4. PMID: 19881175.
31. Shadnia S, Mehrpour O, Soltaninejad K. A simplified acute physiology score in the prediction of acute aluminum phosphide poisoning outcome. Indian J Med Sci. 2010;64(12):532-9. PMID: 21258160.
32. Mostafazadeh B, Pajoumand A, Farzaneh E, Aghabiklooei A, Rasouli MR. Blood levels of methemoglobin in patients with aluminum phosphide poisoning and its correlation with patient's outcome. J Med Toxicol. 2011;7(1):40-3. PMID: 21057909.
33. Hosseinian A, Pakravan N, Rafiei A, Feyzbakhsh SM. Aluminum phosphide poisoning known as rice tablet: A common toxicity in North Iran. Indian J Med Sci. 2011;65(4):143-50. PMID: 23250344.
34. Sharma A, Dishant, Gupta V, Kaushik JS, Mittal K. Aluminum phosphide (celphos) poisoning in children: A 5-year experience in a tertiary care hospital from northern India. Indian J Crit Care Med. 2014;18(1):33-6. PMID: 24550611.
35. Agarwal A, Robo R, Jain N, Gutch M, Consil S, Kumar S. Oxidative stress determined through the levels of antioxidant enzymes and the effect of N-acetylcysteine in aluminum phosphide poisoning. Indian J Crit Care Med. 2014;18(10):666-71. PMID: 25316977.