On 2017 Jun 12, Maria Sammartino commented:

"The interaction of the electron beam, emitted by the gun, with the sample induces an excitation, energy is lost and a single X-ray is emitted that is characteristic of the element hit." A very bad description of the EDS! Anyway the author Gatti A.M. improved her knowledge on the subject; really in one of her oldest article ( Liver and kidney foreign bodies granulomatosis in a patient with malocclusion, bruxism, and worn dental prostheses.By: Ballestri, M; Baraldi, A; Gatti, AM; et al. GASTROENTEROLOGY Volume: 121 Issue: 5 Pages: 1234-1238 Published: NOV 2001)she defined the X-ray microprobe "radiograph microprobe" May be due to the scarce knowlege of the EDS mechanism, that imply an elemental analysis, the error usual in almost all the articles by Gatti is to state that what she find in the sample are non-biodegradable metals (The particles detected showed to contain highly-reactive, non-biocompatible and non-biodegradable metals. It is not specified to which of the white particles the spectra in fig. 1 refer. In my opinion, from the SEM images of the same figure, i.e. at such magnitude, it is almost impossible to measure the particles dimension. Further, even if the total surface occupied by the sample on the acetate filter is not declared, in my opinion, it is anyway almost impossible to count all the particles in a reasonable time; as an example the SEM image show an area of about 50x50 micrometers; how many images they have to acquire to cover a 10x10 mm area? The PCA is at all not explained, and not correctly graphicated. First of all the graph of the two Principal Components must be isometric and squared; the graph of the Loadings lacks and, looking at the data reported in tables and hystogram, it is unclear what they are; the first two components account for less than 47% of the total variance and the graph of the % variance as a function of the components lacks

On 2017 Feb 01, Davide Radice commented:

Visani G et al, compared the occurrence of nanoparticles and aggregates in the peripheral blood samples of AML patients and healthy subjects in a matched case-control study and based on their findings they argue that nanoparticles could lead to AML (that is: nanoparticles could be a risk factor for AML). However I found a number of issues regarding both the design and the statistical analysis. Here in brief the most relevant ones:

1. apart from the few number of subjects included they say that AML patients were matched with healthy subjects, however they do not specify which confounders they matched and controlled for (Table 1 only describes the characteristics of the AML patients but not those of the matched healthy subjects)

2. the primary analysis compares the average particles and aggregates counts, element by element (Table 4) through a series of two-sample t-tests: it should be kept in mind that

   a) the t-test is not suitable for count data because counts violate the underlying assumptions, thus any inference based on the significance of the t-test must be considered as possibly wrong

   b) despite they call each t-test as ‘independent’ because they consider controls as independent of cases, the individual tests are not independent of one another due to the fact that they all were conducted on the same sample of subjects. This is a well-known additional issue called ‘the multiple comparison problem’ they would have to take into account even if they used a more appropriate non-parametric alternative to the t-test. For example taking the 19 raw p-values in Table 4 it can be shown that after adjusting for the multiplicity using the Holm method [1], the only statistically significant comparisons are those regarding the average counts for Al (p = 0.019) and Ca (p = 0.019).

Moreover a statistically significant difference for the average of the counts between AML and healthy subjects it does not imply that aggregates and particles can be considered as possible risk factors, as can not be inferred in general as a risk factor any other variable just on the basis of the significance of the difference between two means (think for a while to a significant difference between the average shoes size, observed by chance). To conclude correctly about a risk you don’t compare two means, you must estimate and test the risks. The authors they should have properly analyze their data using a conditional logistic regression model [2]

Taking the data in Table 2 and assuming that each sample in a row shows the counts of a properly matched case-control for the unspecified confounders, I ran the appropriate statistical analysis (SAS 9.3) as described above. Taking the controls as reference the (multivariable) conditional logistic regression analysis clearly shows that neither aggregates nor particles are significant risk factors for AML. Here the results:

• particles : OR = 1.19 (95% CI: 0.83,1.69) p = 0.34
• aggregates: OR = 0.58 (95% CI: 0.09,3.83) p = 0.57

Legenda: OR = Odds Ratio, CI = Confidence Interval

[1] Holm, S. (1979). "A simple sequentially rejective multiple test procedure". Scandinavian Journal of Statistics. 6 (2): 65–70

[2] Breslow, N. E., et al. (1978). "Estimation of multiple relative risk functions in matched case-control studies.". American Journal of Epidemiology. 108 (4): 299–307

On 2017 Feb 01, Davide Radice commented:

Visani G et al, compared the occurrence of nanoparticles and aggregates in the peripheral blood samples of AML patients and healthy subjects in a matched case-control study and based on their findings they argue that nanoparticles could lead to AML (that is: nanoparticles could be a risk factor for AML). However I found a number of issues regarding both the design and the statistical analysis. Here in brief the most relevant ones:

1. apart from the few number of subjects included they say that AML patients were matched with healthy subjects, however they do not specify which confounders they matched and controlled for (Table 1 only describes the characteristics of the AML patients but not those of the matched healthy subjects)

2. the primary analysis compares the average particles and aggregates counts, element by element (Table 4) through a series of two-sample t-tests: it should be kept in mind that

   a) the t-test is not suitable for count data because counts violate the underlying assumptions, thus any inference based on the significance of the t-test must be considered as possibly wrong

   b) despite they call each t-test as ‘independent’ because they consider controls as independent of cases, the individual tests are not independent of one another due to the fact that they all were conducted on the same sample of subjects. This is a well-known additional issue called ‘the multiple comparison problem’ they would have to take into account even if they used a more appropriate non-parametric alternative to the t-test. For example taking the 19 raw p-values in Table 4 it can be shown that after adjusting for the multiplicity using the Holm method [1], the only statistically significant comparisons are those regarding the average counts for Al (p = 0.019) and Ca (p = 0.019).

Moreover a statistically significant difference for the average of the counts between AML and healthy subjects it does not imply that aggregates and particles can be considered as possible risk factors, as can not be inferred in general as a risk factor any other variable just on the basis of the significance of the difference between two means (think for a while to a significant difference between the average shoes size, observed by chance). To conclude correctly about a risk you don’t compare two means, you must estimate and test the risks. The authors they should have properly analyze their data using a conditional logistic regression model [2]

Taking the data in Table 2 and assuming that each sample in a row shows the counts of a properly matched case-control for the unspecified confounders, I ran the appropriate statistical analysis (SAS 9.3) as described above. Taking the controls as reference the (multivariable) conditional logistic regression analysis clearly shows that neither aggregates nor particles are significant risk factors for AML. Here the results:

• particles : OR = 1.19 (95% CI: 0.83,1.69) p = 0.34
• aggregates: OR = 0.58 (95% CI: 0.09,3.83) p = 0.57

Legenda: OR = Odds Ratio, CI = Confidence Interval

[1] Holm, S. (1979). "A simple sequentially rejective multiple test procedure". Scandinavian Journal of Statistics. 6 (2): 65–70

[2] Breslow, N. E., et al. (1978). "Estimation of multiple relative risk functions in matched case-control studies.". American Journal of Epidemiology. 108 (4): 299–307

On 2017 Jun 12, Maria Sammartino commented:

"The interaction of the electron beam, emitted by the gun, with the sample induces an excitation, energy is lost and a single X-ray is emitted that is characteristic of the element hit." A very bad description of the EDS! Anyway the author Gatti A.M. improved her knowledge on the subject; really in one of her oldest article ( Liver and kidney foreign bodies granulomatosis in a patient with malocclusion, bruxism, and worn dental prostheses.By: Ballestri, M; Baraldi, A; Gatti, AM; et al. GASTROENTEROLOGY Volume: 121 Issue: 5 Pages: 1234-1238 Published: NOV 2001)she defined the X-ray microprobe "radiograph microprobe" May be due to the scarce knowlege of the EDS mechanism, that imply an elemental analysis, the error usual in almost all the articles by Gatti is to state that what she find in the sample are non-biodegradable metals (The particles detected showed to contain highly-reactive, non-biocompatible and non-biodegradable metals. It is not specified to which of the white particles the spectra in fig. 1 refer. In my opinion, from the SEM images of the same figure, i.e. at such magnitude, it is almost impossible to measure the particles dimension. Further, even if the total surface occupied by the sample on the acetate filter is not declared, in my opinion, it is anyway almost impossible to count all the particles in a reasonable time; as an example the SEM image show an area of about 50x50 micrometers; how many images they have to acquire to cover a 10x10 mm area? The PCA is at all not explained, and not correctly graphicated. First of all the graph of the two Principal Components must be isometric and squared; the graph of the Loadings lacks and, looking at the data reported in tables and hystogram, it is unclear what they are; the first two components account for less than 47% of the total variance and the graph of the % variance as a function of the components lacks