Inter-rater reliability of malaria parasite counts and comparison of methods ↔ Inter-observer agreement according to malaria parasite density

LATER¶3-7

Blood film examinations are crucial not only to distinguish parasitaemic from aparasitaemic children, but also to determine the parasite species and their density in the bloodstream. Thus, a correct reading will reduce misclassification bias and yield accurate effect measures. * Correspondence: mounkaila_abdou@hotmail.com; okd@icermali.org 11512 Potomac Ave SE, Washington DC, USA 2Mali-Tulane Tropical Medicine Research Center, Malaria Research and Training Center, Faculty of Medicine, Pharmacy and Odontostomatology, University of Sciences, Techniques and Technology, P.O. Box 1805, Bamako, Mali © 2013 Billo et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Billo et al. Malaria Journal 2013, 12:335 http://www.malariajournal.com/content/12/1/335 The objective of this study was to assess the reproducibility of the results of thick blood smears obtained from a cohort of children of the village of Bancoumana, Mali (West Africa) by re-examining ~10% of the slides. Methods The blood films were collected and prepared with the approval of both the Research Ethics Committees of the Faculty of Medicine, Pharmacy and Odontostomatology of the University of Bamako, Mali and Tulane University, New Orleans, USA. Study participants were enrolled yearly in Bancoumana village, mainly during the transmission season (June to October). All children aged up to ten years were included in the study. The malaria research and training centre has maintained a field laboratory in Bancoumana since June 1993. This village is located within a narrow riverine valley that has an area of approximately 10 sq km. The village itself consists of approximately 10,000 individuals living in 200 houses. Malaria occurs throughout the year with an average monthly prevalence of approximately 50%, with an intense seasonal transmission from June to November [3].

LATER¶5-9

© 2013 Billo et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Billo et al. Malaria Journal 2013, 12:335 http://www.malariajournal.com/content/12/1/335 The objective of this study was to assess the reproducibility of the results of thick blood smears obtained from a cohort of children of the village of Bancoumana, Mali (West Africa) by re-examining ~10% of the slides. Methods The blood films were collected and prepared with the approval of both the Research Ethics Committees of the Faculty of Medicine, Pharmacy and Odontostomatology of the University of Bamako, Mali and Tulane University, New Orleans, USA. Study participants were enrolled yearly in Bancoumana village, mainly during the transmission season (June to October). All children aged up to ten years were included in the study. The malaria research and training centre has maintained a field laboratory in Bancoumana since June 1993. This village is located within a narrow riverine valley that has an area of approximately 10 sq km. The village itself consists of approximately 10,000 individuals living in 200 houses. Malaria occurs throughout the year with an average monthly prevalence of approximately 50%, with an intense seasonal transmission from June to November [3]. Thick smears were stained with 3% Giemsa (Sigma, St Louis, MO, USA) in phosphate buffer (pH 7.0) and examined using oil immersion magnification (1,000 X). The blood smears obtained from the persistently negative children in June, August and October 1996 and March 1997 were systematically re-examined. Also, all negative slides during the four consecutive crosssectional surveys (June 1997 to February 1998) were reexamined. In addition, a stratified random sample (10%) proportional to the following parasite density classes 1–100, 101–5,000, and 5001 and over was taken from the slides collected in June, August and October 1996 (see Table 1). Each slide was examined under oilimmersion (100 ×) until the microscopist had counted the number of asexual parasites (trophozoites) in fields containing 300 or more white blood cells. Parasite counts were estimated by multiplying the number of asexual parasites per 300 white cells by 25 (based on an average white blood cell count of 7,500 per μl) and expressed as the number of parasites per μl. At least

LATER¶17-21

Collapsing the data in a 2×2 table according to the presence or absence of P. falciparum parasite in a thick blood smear gives a Cohen’s κ of 0.7179 (p<0.0001; 95% CI [0.6737, 0.7621]), indicating a excellent reproducibility between the first and the second readings. Figure 1 shows the concordance correlation coefficient [5] computed on the logarithm transformation of the parasite density using William’s method [lnpf = (pf+1)]. This correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [9] (Figure 2). Discussion Currently, in many African countries, the accepted diagnostic technique for malaria is the examination of stained blood films under the oil immersion lens of the microscope. Serology and molecular technique play a part in epidemiology and in various special investigations [10,11]. Light microscopy has a central role in parasite identification and quantification and remains the main method of parasite-based diagnosis in clinic and hospital settings. Thick blood films allow a rapid examination of a relatively large volume of blood, enabling the detection of even scanty parasitaemia of all blood parasites. A wellprepared thick blood film gives more than a ten-fold increase in sensitivity over thin films [12]. Malaria prevalence is decreasing in many African countries. Therefore, the ability to identify all parasites becomes increasingly important. Good quality microscopy conducted by skilled technicians with capacity to manage appropriate quality control, and the currently available rapid diagnosis test (RDT), requiring less training than microscopy, are generally adequate for diagnosis in people who have acute malaria [12]. However, there are issues to be addressed with both procedures. Ensuring the quality of microscopy used for routine diagnosis has often proved difficult as the sensitivity and specificity of routine microscopy is significantly lower when compared with that of qualified microscopists based in reference laboratories [13]. This underlies the need for good training in microscopy for staff in remote areas. The choice of routine diagnosis of malaria in areas of low parasitaemia is microscopy, which is technically more difficult but is better for species identification and for estimating parasite densities, or diagnosis with the user-friendly RDT, which gives a positive or negative result (but not a measure of the density of parasites) and is not good for detecting Plasmodium vivax and the other non-falciparum parasites. Total 608 105 219 932 First Readings Second Readings 0 1 2 3 4 5 6 0 1 2 3 4 5 6 Figure 1 Transformed parasite counts of first and second readings, with line of perfect concordance and regression line using all data (932 slides).

LATER¶17-21

Collapsing the data in a 2×2 table according to the presence or absence of P. falciparum parasite in a thick blood smear gives a Cohen’s κ of 0.7179 (p<0.0001; 95% CI [0.6737, 0.7621]), indicating a excellent reproducibility between the first and the second readings. Figure 1 shows the concordance correlation coefficient [5] computed on the logarithm transformation of the parasite density using William’s method [lnpf = (pf+1)]. This correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [9] (Figure 2). Discussion Currently, in many African countries, the accepted diagnostic technique for malaria is the examination of stained blood films under the oil immersion lens of the microscope. Serology and molecular technique play a part in epidemiology and in various special investigations [10,11]. Light microscopy has a central role in parasite identification and quantification and remains the main method of parasite-based diagnosis in clinic and hospital settings. Thick blood films allow a rapid examination of a relatively large volume of blood, enabling the detection of even scanty parasitaemia of all blood parasites. A wellprepared thick blood film gives more than a ten-fold increase in sensitivity over thin films [12]. Malaria prevalence is decreasing in many African countries. Therefore, the ability to identify all parasites becomes increasingly important. Good quality microscopy conducted by skilled technicians with capacity to manage appropriate quality control, and the currently available rapid diagnosis test (RDT), requiring less training than microscopy, are generally adequate for diagnosis in people who have acute malaria [12]. However, there are issues to be addressed with both procedures. Ensuring the quality of microscopy used for routine diagnosis has often proved difficult as the sensitivity and specificity of routine microscopy is significantly lower when compared with that of qualified microscopists based in reference laboratories [13]. This underlies the need for good training in microscopy for staff in remote areas. The choice of routine diagnosis of malaria in areas of low parasitaemia is microscopy, which is technically more difficult but is better for species identification and for estimating parasite densities, or diagnosis with the user-friendly RDT, which gives a positive or negative result (but not a measure of the density of parasites) and is not good for detecting Plasmodium vivax and the other non-falciparum parasites. Total 608 105 219 932 First Readings Second Readings 0 1 2 3 4 5 6 0 1 2 3 4 5 6 Figure 1 Transformed parasite counts of first and second readings, with line of perfect concordance and regression line using all data (932 slides).

LATER¶20-24

Malaria prevalence is decreasing in many African countries. Therefore, the ability to identify all parasites becomes increasingly important. Good quality microscopy conducted by skilled technicians with capacity to manage appropriate quality control, and the currently available rapid diagnosis test (RDT), requiring less training than microscopy, are generally adequate for diagnosis in people who have acute malaria [12]. However, there are issues to be addressed with both procedures. Ensuring the quality of microscopy used for routine diagnosis has often proved difficult as the sensitivity and specificity of routine microscopy is significantly lower when compared with that of qualified microscopists based in reference laboratories [13]. This underlies the need for good training in microscopy for staff in remote areas. The choice of routine diagnosis of malaria in areas of low parasitaemia is microscopy, which is technically more difficult but is better for species identification and for estimating parasite densities, or diagnosis with the user-friendly RDT, which gives a positive or negative result (but not a measure of the density of parasites) and is not good for detecting Plasmodium vivax and the other non-falciparum parasites. Total 608 105 219 932 First Readings Second Readings 0 1 2 3 4 5 6 0 1 2 3 4 5 6 Figure 1 Transformed parasite counts of first and second readings, with line of perfect concordance and regression line using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 3 of 6 http://www.malariajournal.com/content/12/1/335 Parasite density estimation is highly valuable for the clinician, as it is an important determinant of treatment schedules for P. falciparum. If parasite density exceeds 10% in P. falciparum, exchange transfusion may be indicated [14,15]. A variety of studies have clearly demonstrated that microscopic diagnosis of malaria can vary greatly in its accuracy, particularly at low parasitaemia rates [12,16,17]. This variation in specificity and sensitivity is routinely observed in clinical settings, where a high proportion of reporting patients are parasitaemic and parasite densities are relatively high. In this study, the weighted kappa statistic, widely used as a chancecorrected measure for nominal agreement, showed excellent inter-observer agreement (κw=0.7926; 95% CI [0.7588, 0.8263]; p=0.01). Inter-observer agreement for slides read as negative by one observer, or as containing more than 100 parasites per μl was excellent: 97% (493/506) and 92% (145/158), respectively. Dowling and Shute compared parasite counts obtained by examination of thin and thick smears and conducted that parasite losses of 60 to 90% occurred with thick films, whereas since thin films are fixed after drying and before staining, they assumed no significant loss of parasites during staining [18]. In a series of parasite dilutions, studies have found that thick films tended to measure parasite densities around one log lower than the number calculated to be in the dilution and this did not vary by microscopist [13,19]. O’Meara et al. [19] have shown that parasitaemia from the thick smear averaged 10% lower than the total mean (p = 0.001) and they have also shown that white blood cells were much less uniformly distributed that the parasites. They also confirmed that up to 60% of parasites were obscured in the thick film or lost during the process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities.

LATER¶20-24

Malaria prevalence is decreasing in many African countries. Therefore, the ability to identify all parasites becomes increasingly important. Good quality microscopy conducted by skilled technicians with capacity to manage appropriate quality control, and the currently available rapid diagnosis test (RDT), requiring less training than microscopy, are generally adequate for diagnosis in people who have acute malaria [12]. However, there are issues to be addressed with both procedures. Ensuring the quality of microscopy used for routine diagnosis has often proved difficult as the sensitivity and specificity of routine microscopy is significantly lower when compared with that of qualified microscopists based in reference laboratories [13]. This underlies the need for good training in microscopy for staff in remote areas. The choice of routine diagnosis of malaria in areas of low parasitaemia is microscopy, which is technically more difficult but is better for species identification and for estimating parasite densities, or diagnosis with the user-friendly RDT, which gives a positive or negative result (but not a measure of the density of parasites) and is not good for detecting Plasmodium vivax and the other non-falciparum parasites. Total 608 105 219 932 First Readings Second Readings 0 1 2 3 4 5 6 0 1 2 3 4 5 6 Figure 1 Transformed parasite counts of first and second readings, with line of perfect concordance and regression line using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 3 of 6 http://www.malariajournal.com/content/12/1/335 Parasite density estimation is highly valuable for the clinician, as it is an important determinant of treatment schedules for P. falciparum. If parasite density exceeds 10% in P. falciparum, exchange transfusion may be indicated [14,15]. A variety of studies have clearly demonstrated that microscopic diagnosis of malaria can vary greatly in its accuracy, particularly at low parasitaemia rates [12,16,17]. This variation in specificity and sensitivity is routinely observed in clinical settings, where a high proportion of reporting patients are parasitaemic and parasite densities are relatively high. In this study, the weighted kappa statistic, widely used as a chancecorrected measure for nominal agreement, showed excellent inter-observer agreement (κw=0.7926; 95% CI [0.7588, 0.8263]; p=0.01). Inter-observer agreement for slides read as negative by one observer, or as containing more than 100 parasites per μl was excellent: 97% (493/506) and 92% (145/158), respectively. Dowling and Shute compared parasite counts obtained by examination of thin and thick smears and conducted that parasite losses of 60 to 90% occurred with thick films, whereas since thin films are fixed after drying and before staining, they assumed no significant loss of parasites during staining [18]. In a series of parasite dilutions, studies have found that thick films tended to measure parasite densities around one log lower than the number calculated to be in the dilution and this did not vary by microscopist [13,19]. O’Meara et al. [19] have shown that parasitaemia from the thick smear averaged 10% lower than the total mean (p = 0.001) and they have also shown that white blood cells were much less uniformly distributed that the parasites. They also confirmed that up to 60% of parasites were obscured in the thick film or lost during the process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities.

LATER¶21-25

Total 608 105 219 932 First Readings Second Readings 0 1 2 3 4 5 6 0 1 2 3 4 5 6 Figure 1 Transformed parasite counts of first and second readings, with line of perfect concordance and regression line using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 3 of 6 http://www.malariajournal.com/content/12/1/335 Parasite density estimation is highly valuable for the clinician, as it is an important determinant of treatment schedules for P. falciparum. If parasite density exceeds 10% in P. falciparum, exchange transfusion may be indicated [14,15]. A variety of studies have clearly demonstrated that microscopic diagnosis of malaria can vary greatly in its accuracy, particularly at low parasitaemia rates [12,16,17]. This variation in specificity and sensitivity is routinely observed in clinical settings, where a high proportion of reporting patients are parasitaemic and parasite densities are relatively high. In this study, the weighted kappa statistic, widely used as a chancecorrected measure for nominal agreement, showed excellent inter-observer agreement (κw=0.7926; 95% CI [0.7588, 0.8263]; p=0.01). Inter-observer agreement for slides read as negative by one observer, or as containing more than 100 parasites per μl was excellent: 97% (493/506) and 92% (145/158), respectively. Dowling and Shute compared parasite counts obtained by examination of thin and thick smears and conducted that parasite losses of 60 to 90% occurred with thick films, whereas since thin films are fixed after drying and before staining, they assumed no significant loss of parasites during staining [18]. In a series of parasite dilutions, studies have found that thick films tended to measure parasite densities around one log lower than the number calculated to be in the dilution and this did not vary by microscopist [13,19]. O’Meara et al. [19] have shown that parasitaemia from the thick smear averaged 10% lower than the total mean (p = 0.001) and they have also shown that white blood cells were much less uniformly distributed that the parasites. They also confirmed that up to 60% of parasites were obscured in the thick film or lost during the process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities. In contrast, the inter-observer agreement for slides read by one observer as containing 1–100 parasites/μl was poor, 36% (96/268). The concordance correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [10] [Figure 2]. Greenwood and Armstrong [20] have suggested that variation in parasite density depends in variability in the volume of blood used to prepare thick films being less than the variability in white blood cell count in the population they studied.

LATER¶21-26

Total 608 105 219 932 First Readings Second Readings 0 1 2 3 4 5 6 0 1 2 3 4 5 6 Figure 1 Transformed parasite counts of first and second readings, with line of perfect concordance and regression line using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 3 of 6 http://www.malariajournal.com/content/12/1/335 Parasite density estimation is highly valuable for the clinician, as it is an important determinant of treatment schedules for P. falciparum. If parasite density exceeds 10% in P. falciparum, exchange transfusion may be indicated [14,15]. A variety of studies have clearly demonstrated that microscopic diagnosis of malaria can vary greatly in its accuracy, particularly at low parasitaemia rates [12,16,17]. This variation in specificity and sensitivity is routinely observed in clinical settings, where a high proportion of reporting patients are parasitaemic and parasite densities are relatively high. In this study, the weighted kappa statistic, widely used as a chancecorrected measure for nominal agreement, showed excellent inter-observer agreement (κw=0.7926; 95% CI [0.7588, 0.8263]; p=0.01). Inter-observer agreement for slides read as negative by one observer, or as containing more than 100 parasites per μl was excellent: 97% (493/506) and 92% (145/158), respectively. Dowling and Shute compared parasite counts obtained by examination of thin and thick smears and conducted that parasite losses of 60 to 90% occurred with thick films, whereas since thin films are fixed after drying and before staining, they assumed no significant loss of parasites during staining [18]. In a series of parasite dilutions, studies have found that thick films tended to measure parasite densities around one log lower than the number calculated to be in the dilution and this did not vary by microscopist [13,19]. O’Meara et al. [19] have shown that parasitaemia from the thick smear averaged 10% lower than the total mean (p = 0.001) and they have also shown that white blood cells were much less uniformly distributed that the parasites. They also confirmed that up to 60% of parasites were obscured in the thick film or lost during the process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities. In contrast, the inter-observer agreement for slides read by one observer as containing 1–100 parasites/μl was poor, 36% (96/268). The concordance correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [10] [Figure 2]. Greenwood and Armstrong [20] have suggested that variation in parasite density depends in variability in the volume of blood used to prepare thick films being less than the variability in white blood cell count in the population they studied. When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers

LATER¶23-27

In a series of parasite dilutions, studies have found that thick films tended to measure parasite densities around one log lower than the number calculated to be in the dilution and this did not vary by microscopist [13,19]. O’Meara et al. [19] have shown that parasitaemia from the thick smear averaged 10% lower than the total mean (p = 0.001) and they have also shown that white blood cells were much less uniformly distributed that the parasites. They also confirmed that up to 60% of parasites were obscured in the thick film or lost during the process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities. In contrast, the inter-observer agreement for slides read by one observer as containing 1–100 parasites/μl was poor, 36% (96/268). The concordance correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [10] [Figure 2]. Greenwood and Armstrong [20] have suggested that variation in parasite density depends in variability in the volume of blood used to prepare thick films being less than the variability in white blood cell count in the population they studied. When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides).

LATER¶24-28

process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities. In contrast, the inter-observer agreement for slides read by one observer as containing 1–100 parasites/μl was poor, 36% (96/268). The concordance correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [10] [Figure 2]. Greenwood and Armstrong [20] have suggested that variation in parasite density depends in variability in the volume of blood used to prepare thick films being less than the variability in white blood cell count in the population they studied. When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider.

LATER¶24-28

process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities. In contrast, the inter-observer agreement for slides read by one observer as containing 1–100 parasites/μl was poor, 36% (96/268). The concordance correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [10] [Figure 2]. Greenwood and Armstrong [20] have suggested that variation in parasite density depends in variability in the volume of blood used to prepare thick films being less than the variability in white blood cell count in the population they studied. When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider.

LATER¶24-28

process of red cell lyses and parasite staining. In this study, agreement was compared between two highly qualified microscopists according to parasite densities. In contrast, the inter-observer agreement for slides read by one observer as containing 1–100 parasites/μl was poor, 36% (96/268). The concordance correlation co-efficient [5] was 0.835, 95% CI (0.816-0.855) using logarithm-transformed parasite counts, and yielded a regression line with near-perfect concordance between the first and the second readings: an average difference of −0.088 ±0.474 [10] [Figure 2]. Greenwood and Armstrong [20] have suggested that variation in parasite density depends in variability in the volume of blood used to prepare thick films being less than the variability in white blood cell count in the population they studied. When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider.

LATER¶26-30

When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider. Since most elimination efforts will need to deal with both low parasitaemia and non-falciparum species, diagnosis becomes a major challenge for elimination programmes. Bowers et al. have shown differences between methods using the same microscopy staff, but reader technique itself clearly contributes to the accuracy of parasitaemia estimates [26]. Although the propensity of a gametocyte carrier to transmit infection is related to the density of gametocytaemia, individuals with very low gametocyte numbers can still transmit malaria infection and can be an important part of the reservoir of infection. Thus, elimination programmes will need to detect and treat all potential transmitters of infection with a more sensitive detection test. The slide readers in this study were all experienced malaria microscopists and the results may be different with less experienced readers. In the light of this and under low parasite prevalence, low parasite rates, and inadequate equipment conditions, for any parasite density less than 100 parasites/μl at least two experienced microscopists should blind read the slide. Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in

LATER¶26-30

When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider. Since most elimination efforts will need to deal with both low parasitaemia and non-falciparum species, diagnosis becomes a major challenge for elimination programmes. Bowers et al. have shown differences between methods using the same microscopy staff, but reader technique itself clearly contributes to the accuracy of parasitaemia estimates [26]. Although the propensity of a gametocyte carrier to transmit infection is related to the density of gametocytaemia, individuals with very low gametocyte numbers can still transmit malaria infection and can be an important part of the reservoir of infection. Thus, elimination programmes will need to detect and treat all potential transmitters of infection with a more sensitive detection test. The slide readers in this study were all experienced malaria microscopists and the results may be different with less experienced readers. In the light of this and under low parasite prevalence, low parasite rates, and inadequate equipment conditions, for any parasite density less than 100 parasites/μl at least two experienced microscopists should blind read the slide. Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in

LATER¶26-30

When two parasite counts for the same slide were compared, Killian et al. found considerable variability, with one reading being 0.12 to ten times the other [21]. They examined inter-rater variability in the results of malaria microscopy in epidemiological studies using 711 thick blood films re-read by four experienced microscopists. They also calculated parasite density by counting the number of trophozoites in 100 oil immersion fields and multiplying by four to give parasites per microlitre, assuming a blood volume of approximately 0.25 μl per 100 microscope fields. There was significantly less variability at parasite densities above 500/μl, 0.2 to 3.6 times. Overall, for variation between readers, O’Meara et al. stated that discrepancies in parasite densities reported by experienced clinic microscopists decreased with increasing mean density and trends were similar for P. falciparum and for P. vivax when they were considered separately [22]. When agreement between readers Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider. Since most elimination efforts will need to deal with both low parasitaemia and non-falciparum species, diagnosis becomes a major challenge for elimination programmes. Bowers et al. have shown differences between methods using the same microscopy staff, but reader technique itself clearly contributes to the accuracy of parasitaemia estimates [26]. Although the propensity of a gametocyte carrier to transmit infection is related to the density of gametocytaemia, individuals with very low gametocyte numbers can still transmit malaria infection and can be an important part of the reservoir of infection. Thus, elimination programmes will need to detect and treat all potential transmitters of infection with a more sensitive detection test. The slide readers in this study were all experienced malaria microscopists and the results may be different with less experienced readers. In the light of this and under low parasite prevalence, low parasite rates, and inadequate equipment conditions, for any parasite density less than 100 parasites/μl at least two experienced microscopists should blind read the slide. Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in

LATER¶27-31

Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider. Since most elimination efforts will need to deal with both low parasitaemia and non-falciparum species, diagnosis becomes a major challenge for elimination programmes. Bowers et al. have shown differences between methods using the same microscopy staff, but reader technique itself clearly contributes to the accuracy of parasitaemia estimates [26]. Although the propensity of a gametocyte carrier to transmit infection is related to the density of gametocytaemia, individuals with very low gametocyte numbers can still transmit malaria infection and can be an important part of the reservoir of infection. Thus, elimination programmes will need to detect and treat all potential transmitters of infection with a more sensitive detection test. The slide readers in this study were all experienced malaria microscopists and the results may be different with less experienced readers. In the light of this and under low parasite prevalence, low parasite rates, and inadequate equipment conditions, for any parasite density less than 100 parasites/μl at least two experienced microscopists should blind read the slide. Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in malaria elimination phase as most African countries are observing a decrease in malaria prevalence. Additional file Additional file 1: Between-readers variation in asexual parasites counts. Title: Variation between readers in asexual parasites counts using Alexander et al. (2010) methods. Description: When back-transformed to the original parasite density count, the limits with agreements increase with parasite density, and are much wider.

LATER¶27-31

Figure 2 Difference against average of transformed parasite counts of first and second readings, with 95% limits of agreement using all data (932 slides). Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider. Since most elimination efforts will need to deal with both low parasitaemia and non-falciparum species, diagnosis becomes a major challenge for elimination programmes. Bowers et al. have shown differences between methods using the same microscopy staff, but reader technique itself clearly contributes to the accuracy of parasitaemia estimates [26]. Although the propensity of a gametocyte carrier to transmit infection is related to the density of gametocytaemia, individuals with very low gametocyte numbers can still transmit malaria infection and can be an important part of the reservoir of infection. Thus, elimination programmes will need to detect and treat all potential transmitters of infection with a more sensitive detection test. The slide readers in this study were all experienced malaria microscopists and the results may be different with less experienced readers. In the light of this and under low parasite prevalence, low parasite rates, and inadequate equipment conditions, for any parasite density less than 100 parasites/μl at least two experienced microscopists should blind read the slide. Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in malaria elimination phase as most African countries are observing a decrease in malaria prevalence. Additional file Additional file 1: Between-readers variation in asexual parasites counts. Title: Variation between readers in asexual parasites counts using Alexander et al. (2010) methods. Description: When back-transformed to the original parasite density count, the limits with agreements increase with parasite density, and are much wider.

LATER¶28-32

Billo et al. Malaria Journal 2013, 12:335 Page 4 of 6 http://www.malariajournal.com/content/12/1/335 is required, it is important to apply an identical technique which seems to be more important than increasing the number of microscope fields read [22]. In another study, these authors found a significant inverse correlation between discrepancy among microscopists and mean parasite density [23]. Furthermore, they suggested that random chance in the selection of fields to examine may play a large part in reader discrepancy, especially with low parasitaemia. In a recent review, Makler et al. concluded that factors such as undertraining of microscopists, lack of microscopes and staining materials, and processing and reading large numbers of blood smears, dramatically increased the range for error [24]. Using the method described by Alexander et al. [25], similar findings were observed in the present study (see Additional file 1). When back-transformed to the original count, the limits with agreements increase with parasite density, and are much wider. Since most elimination efforts will need to deal with both low parasitaemia and non-falciparum species, diagnosis becomes a major challenge for elimination programmes. Bowers et al. have shown differences between methods using the same microscopy staff, but reader technique itself clearly contributes to the accuracy of parasitaemia estimates [26]. Although the propensity of a gametocyte carrier to transmit infection is related to the density of gametocytaemia, individuals with very low gametocyte numbers can still transmit malaria infection and can be an important part of the reservoir of infection. Thus, elimination programmes will need to detect and treat all potential transmitters of infection with a more sensitive detection test. The slide readers in this study were all experienced malaria microscopists and the results may be different with less experienced readers. In the light of this and under low parasite prevalence, low parasite rates, and inadequate equipment conditions, for any parasite density less than 100 parasites/μl at least two experienced microscopists should blind read the slide. Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in malaria elimination phase as most African countries are observing a decrease in malaria prevalence. Additional file Additional file 1: Between-readers variation in asexual parasites counts. Title: Variation between readers in asexual parasites counts using Alexander et al. (2010) methods. Description: When back-transformed to the original parasite density count, the limits with agreements increase with parasite density, and are much wider. Competing interests The authors have declared that they have no competing interests. Authors’ contributions AD and MD were the reference microscopists; OKD was the trainer in malaria slide microscopy and served as the third reference microscopist for the entire Mali-Tulane TMRC grant. AD, MD and OKD participated in drafting the manuscript and supervised the collection of samples. MAB, MD, ESJ, JCR, OKD and DJK conceived the study and participated in its design. OKD and MAB supervised all aspects of the study carried out in Bancoumana and the drafting of the manuscript. MAB, JCR and ESJ performed all the statistical analysis for the study. All authors read and approved the final manuscript.

LATER¶30-33

Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in malaria elimination phase as most African countries are observing a decrease in malaria prevalence. Additional file Additional file 1: Between-readers variation in asexual parasites counts. Title: Variation between readers in asexual parasites counts using Alexander et al. (2010) methods. Description: When back-transformed to the original parasite density count, the limits with agreements increase with parasite density, and are much wider. Competing interests The authors have declared that they have no competing interests. Authors’ contributions AD and MD were the reference microscopists; OKD was the trainer in malaria slide microscopy and served as the third reference microscopist for the entire Mali-Tulane TMRC grant. AD, MD and OKD participated in drafting the manuscript and supervised the collection of samples. MAB, MD, ESJ, JCR, OKD and DJK conceived the study and participated in its design. OKD and MAB supervised all aspects of the study carried out in Bancoumana and the drafting of the manuscript. MAB, JCR and ESJ performed all the statistical analysis for the study. All authors read and approved the final manuscript. Acknowledgements The authors are grateful to the population of Bancoumana and the TMRCMRTC-DEAP field and laboratory teams for the quality of data collection on the field sites at Bancoumana, and to the Dean Pr Issa Traoré of the Ecole Nationale de Médecine et de Pharmacie for his full support of the MaliTulane TMRC activities. The project was funded by the Mali-Tulane Tropical Research Centre, funded by the NIH-extramural research programme (NIAID P50 AI 39469) MAB’s PhD activities in Mali were supported by the WHO/TDR through the Special Programme for Research and Training in Tropical Diseases, grant ID-930736.

LATER¶30-33

Conclusion Improved means to detect asymptomatic persons with low parasitaemia will be crucial to malaria elimination. These results suggest a high reproducibility for slides reported as negative or as having more than 100 parasites per μl. However, low parasitaemia (<100 parasites/μl are less reproducible and should be re-examined carefully. In addition, a uniform counting protocol should be used and the number of white blood cells counted should be increased in order to improve inter-reader agreement. Until rapid, reproducible and quantitative PCR for malaria is widely available at low cost, microscopy will remain the method of choice for parasite density determination in malaria elimination phase as most African countries are observing a decrease in malaria prevalence. Additional file Additional file 1: Between-readers variation in asexual parasites counts. Title: Variation between readers in asexual parasites counts using Alexander et al. (2010) methods. Description: When back-transformed to the original parasite density count, the limits with agreements increase with parasite density, and are much wider. Competing interests The authors have declared that they have no competing interests. Authors’ contributions AD and MD were the reference microscopists; OKD was the trainer in malaria slide microscopy and served as the third reference microscopist for the entire Mali-Tulane TMRC grant. AD, MD and OKD participated in drafting the manuscript and supervised the collection of samples. MAB, MD, ESJ, JCR, OKD and DJK conceived the study and participated in its design. OKD and MAB supervised all aspects of the study carried out in Bancoumana and the drafting of the manuscript. MAB, JCR and ESJ performed all the statistical analysis for the study. All authors read and approved the final manuscript. Acknowledgements The authors are grateful to the population of Bancoumana and the TMRCMRTC-DEAP field and laboratory teams for the quality of data collection on the field sites at Bancoumana, and to the Dean Pr Issa Traoré of the Ecole Nationale de Médecine et de Pharmacie for his full support of the MaliTulane TMRC activities. The project was funded by the Mali-Tulane Tropical Research Centre, funded by the NIH-extramural research programme (NIAID P50 AI 39469) MAB’s PhD activities in Mali were supported by the WHO/TDR through the Special Programme for Research and Training in Tropical Diseases, grant ID-930736.

LATER¶31-33

malaria elimination phase as most African countries are observing a decrease in malaria prevalence. Additional file Additional file 1: Between-readers variation in asexual parasites counts. Title: Variation between readers in asexual parasites counts using Alexander et al. (2010) methods. Description: When back-transformed to the original parasite density count, the limits with agreements increase with parasite density, and are much wider. Competing interests The authors have declared that they have no competing interests. Authors’ contributions AD and MD were the reference microscopists; OKD was the trainer in malaria slide microscopy and served as the third reference microscopist for the entire Mali-Tulane TMRC grant. AD, MD and OKD participated in drafting the manuscript and supervised the collection of samples. MAB, MD, ESJ, JCR, OKD and DJK conceived the study and participated in its design. OKD and MAB supervised all aspects of the study carried out in Bancoumana and the drafting of the manuscript. MAB, JCR and ESJ performed all the statistical analysis for the study. All authors read and approved the final manuscript. Acknowledgements The authors are grateful to the population of Bancoumana and the TMRCMRTC-DEAP field and laboratory teams for the quality of data collection on the field sites at Bancoumana, and to the Dean Pr Issa Traoré of the Ecole Nationale de Médecine et de Pharmacie for his full support of the MaliTulane TMRC activities. The project was funded by the Mali-Tulane Tropical Research Centre, funded by the NIH-extramural research programme (NIAID P50 AI 39469) MAB’s PhD activities in Mali were supported by the WHO/TDR through the Special Programme for Research and Training in Tropical Diseases, grant ID-930736.