Midterm 1 Practice

Intro to Analytical Chem, Statistics, Quality Assurance and Calibration Method

when do you calibrate?

when the relationship between analyte concentration and signal/response is unknown

negative bias

mean is below true value (underestimation)

positive value

mean is above true value (overestimation)

low precision results in

difficulty in identifying small differences between data (t-test)

rounding

• @5 round to nearest even digit

• ± answer has same # of digits

• x or / answer has same # of sigfigs

systematic error (determinate)

flaw in method/design, reproducible, can be detected and corrected, skews data in 1 direction, affects accuracy

random error (indeterminate)

uncertainty in measurements that cannot be eliminated, can be reduced by better precision, affects reproducibility, aka variance of data

gross erros/blunders

human errors

uncertainty calculations

• x / use relative uncertainty

• + - use absolute uncertainty

• calibration curve uncertainty use formula

matrix

everything else in sample other than analyte

bias

deviation from truth in data collection, analysis, interpretation, publication, can cause false conclusions

types of bias

• fraud

• selective reporting

• publication bias

• confirmation bias

source of bias

• design

• sampling

• presentation

• data quality

scientific misconduct (2 types)

• fraud (intentional)

• unintentional, erroneous

Intentional scientific misconduct

falsification, fabrication, plagiarism (FFPs)

Unintentional scientific misconduct

• misuse of stats

• image manipulation

• lack of access to primary data

• witholding scientific info

What are some strategies that you can implement in the laboratory to avoid bias?

• have a null hypothesis to alleviate some confirmation bias

• quality assurance

• use blinding and masking techniques also reduces confirmation bias

reproducibility crisis due to

• pressure to publish

• selective reporting

• insufficient replication

• inadequate validation

• low statistical power

most published research is false because

study power is low

study power is dependent on

• sample size

• effect size (response size)

• tested relations

hierarchy in quality of evidence

• lowest expert opinions

• animal studies

• randomized control trials

• meta analysis (results of many studies) highest

reproducibility crisis

where scientists are unable to produce results from published research papers when they attempt to reproduce their studies, many cannot even reproduce their own results → suggests that most published data is false

what are some strategies we can use to improve reproducibility & scientific integrity?

• declaration of conflicts of interests

• data transparency

• adequate replication with enough study power

• eliminating sources of bias in study design

• method validation

• appropriate statistical analysis

biggest source of error in lab 1

vision which is relative person to person to spot endpoint

Precision

is a measure of indeterminate error, amount of variance in replicate measurements

how to quantify precision

• standard deviation

• relative standard deviation/coefficient of variance

• confidence intervals

to increase precision

can have more replicants

standard deviation (s) of sample

• measures how closely data is clustered about the mean

• +- 1s covers 68% of data

• report +- 2s

• larger s, poor precision

• breadth of gaussian distribution of data

sigma

standard deviation of population

mu

mean of population

x bar

mean of sample, central tendency of gaussian distribution

relative standard deviation (RSD)/coefficient of variance (CV)

• s/xbar

• standard deviation as a % of the mean

• RSD < 15% acceptable

• RSD > 20% is poor precision

degrees of freedom

sample size - 1 (for 1 data set)

field blank

environmental analysis, sample that goes through experiment @ site of sampling (see if contaminants added in shipment)

confidence intervals

• range of values in which we are (confidence level %) confident that the true mean lies in

• ± ts/sqrt(n)

• CL is usually 95%

• more replicates (larger n), smaller CI

• use instead of standard deviations cuz 3 replicates not enough for gaussian distribution

n

number of replicates

student’s t test value

statistical tool used for confidence intervals and comparing results from different experiments

• incr dof, decreases t-crit

• incr CL, increase t-crit

reference ranges (for disease vs. healthy states)

• test healthy pop

• find 95% CI → reference range

• if data is outside CI means unhealthy

• used for diagnostic testing

accuracy

• closeness of measured result to true value

• % accuracy

• check for bias in different ways (standard reference materials, blanks, compare with another validated method aka Bland Altman plot, round-robin experiments, spike and recovery test)

% accuracy

• reveals closeness of measured result to true value

• your average/true value

• acceptable 85%-115%

blank sample analysis

run method on sample containing no analyte → should get zero response

round-robin experiment

same experiment is done by different labs/ppl, if the difference in data cannot be explained by random error —> sign of systematic error

sources of systematic error

• sample preparation

• human

• equipment problem

• reporting error

• calculation error

  there are many pre&post analytical factors that can lead to bias in an analytical method

cognitive biases

• human nature (interpret)

• environment, culture, experience

• case specific

s, RSD, CI all assume

normal distribution of data

how to check for outliers

• rank data

• grubbs test

• outher basic statistical tests (visualize)

Grubbs Test

• null hypothesis (there are no outliers in data, all data is random)

• calculate mean

• calculate stdev

• apply formula to all data points: (|sus value-mean|)/s

• if g-calc>g-crit then can reject null hypothesis and claim that data point cannot be explained by random error, and can thus be removed

p value

• 1-CL

• probability for error

types of t-tests

• single-tailed t-test

• 2-tailed t-test

  (both can either be with equal variance or unequal variance)

• paired t-test

F-test

• always done before t-test

• null: the variance between the 2 data sets are equal

• f-calc = s1²/s2²

• if f-calc>f-table, reject null and claim that variance between the data sets are unequal

• alternative to f-test (j/ look at variance and see if its similar in magnitude)

2-tailed t-test

• use to compare means between 2 data sets

• null: 2 means are not significantly different

• calculate means

• calculate stdev

• calculate t-calc

• if t-calc>t-table then can reject null, claim that the means are significantly different

• testing to see if mean is within 5% error total on EITHER side of graph

x - stat

x - calc

x - crit

x- table

full change (FC)

ratio of different mean concentrations

round error to 2-3 sigfigs

round mean accordingly

paired t-test

• compare means from tests applied to identical samples (ex. same test subject)

• if means are not significantly different, can conclude that methods give the same answer

• ex. BICARB muscle fatigue study

variance (s²)

another measure of data’s dispersion around mean

endpoint of titration

• the point at which a color change has been reached indicating a slight excess of titrant has been added to an unknown amount of titrand

• visually detect equivalence point has been passed

equivalence point of titration

point at which moles of analyte is equal to moles of titrant

standard operating procedure (SOP)

states steps taken in an validated analytical protocol: 

standardization titration

titration of a known amount of analyte to determine the

concentration of the titrant

direct translation

titrant is added to analyte until reaction is complete

titration error

the difference between the end point and the equivalence point

single-tailed t-test

• compare mean to a known value (ex. MAC)

• null: the mean and the known value are not significantly different

• FIRST check if the known value is within the upper and lower CI of the mean, if it isn’t reject null and claim they are significantly different, if it does cannot reject null

• calculate t-stat (different formula and t-crit values than 2 tailed t-test)

• if t-stat>t-crit, reject null and claim that mean and known value are significantly different

• testing to see if mean is within 5% error on ONE SIDE OF GRAPH

to compare means from 3+ data sets

• ANOVA

• iterative t-tests (if trying to find max, take largest value and do t-tests with that value and all other values)

pregnancy test

measures hormone released in early stages of fetus development

type 1 error

• false positive

• rejecting true null hypothesis

• ex. pregnancy test positive when not pregnant

• causes: interferences in sample, human erro

type 2 error

• false negative

• accepting a false null hypothesis

• ex. pregnancy test is negative when acc are pregnant

• causes: early measurement, below detect limit

linear regression

• measure the change in 1 variable (compared to another related variable)

• ex. R²

person’s correlation

• measures the change in 2 related variables

• r

parametric test

statistical test that assumes normal distribution of data

non-parametric test

test that doesn’t rely on assumptions about the distribution of data

symmetric data sets

random distribution, equal variance between the two groups

biomodal data set

suggests batch effect (samples tested in different batches cause different results, suggesting non-random variation)

bar graph can mask

• unequal variance between groups

• unequal replicates in each group

• outliers

non-gaussian distribution of data

• can tell if stdev is larger than mean

• report median and IQR (not mean and CI)

• do non-parametric tests

designing an analytical method

• choose scope of analyte (single molecule vs. molecular class)

• know physiochemical properties of analyte to choose a selective method (unique properties can be used to separate & detect analyte)

• identify sample matrix & matric effects in order to determine how to sample prep

• know what the expected analyte conc is (check with method’s detect limit)

• what precision and accuracy is expected

real world sample

• complex

• heterogenous

• do not act like analyte in pure solution

• requires measurement of minor analyte in prescence of major chemical compounds

matrix effects

components of matrix that represent major source of interference

• ex. interferences could co-elute, spike/suppress signal, spectral overlap

analyte

target species to be measured/identified

Thalidomide tradegy due to

drug compound having an enantiomer that is toxic which is found in equal concentration as drug compound, Frances Oldham a hero for stopping people from buying it

sample pretreatments

sample workup done to make sample analyzable (into homogenous solution at appropriate analyte concentration with minimum intereferences)

• ex. homogenization (blending), acid digestion, centrifugation, dilution

• con: can introduce error the more complicated sample preparation is (try to minimize)

method validation

• process of proving that an analytical method is acceptable for its intended purpose (measuring that analyte in that sample)

• assessed according to 8 figures of merit & with calibration methods

8 figures of merit

method specificity, linearity, accuracy, presicion, range, limit of detection (LOD), limit of quantification (LOQ), robustness

specificity/selectivity

• ability of an analytical method to distinguish (measure just) analyte from everything else that might be in sample

• consider all known interferences and determine impact on analyte signal

• perform stress test

• analyze representative samples (that rep entire pop)

• blank sample should get zero response

• ASK: can your method accurately demonstrate that it is measuring just analyte response?

• to improve: do sample prep to remove interferences, change method to a more selective one

examples of selectivity/specificity in class

• ex. mass spectrometry is also very specific because it reports specific fragment mass values

• ex of poor specificity: to test heparin medicine, tested the medicine’s clotting ability, didn’t acc test for analyte —> don’t know what is causing clots (in some batches it was a toxic compound)

stress test

purposefully degrade analyte to see what the byproducts of degradation responses look like

• ascorbic acid is oxidized (looses electrons, loss of H, gain of O)

• iodine is the oxidizing agent (is reduced to iodide)

• iodine is colourless, when no more AA to react with, it reacts with starch

• starch-iodide complex is purple

why is column chromatography more specific than titration?

• titration only get 1 response for entire sample (ex. a/b titration measures bulk a/b capacity of sample)

• column chromatography can identify what compound gave what response (peak)

repeatability

• precision of results when measuring same sample with same method, operator, on same day, in same conditions

• less variance, less trustworthy

intermediate precision/within lab reproducability

precision when measuring analyte from single lab over longer time with different analysts

(between lab) reproducibility

• standardized measurements performed at different laboratories

• important for method to be transferred across multiple labs globally

• depends on context

• more variance, more trustworthy

interference

species that causes a bias in analysis by enhancing/decreasing instrument response to the analyte

total response

analyte sensitvity*concentration of analyte + (interference sensitivty*concentration of interference) should be low so can remove from equation

sensitivity

slope of calibration curve; method’s change in signal in response to change in analyte concentration

selectivity co-eficient

• analyte sensitivity compared to interference sensitivity

• if >100 then method is selective

• ma/mI

spike & recovery analysis

• add known amounts of analyte to a spiked sample

• difference in response between sample and spiked sample should equal the known amount of analyte added

• if not that indicates matrix interference, poor accuracy

Bland Altman Plot

• used to compare results with results from another validated method

• the data points plotted should be within 1-2 standard deviations of the mean from the validated method

• should see random variation

• ±12% is acceptable

calibration methods

• external standard

• internal standard

• standard addition

• only applicable to specific sample