Summary

The invention relates to a method and laboratory equipment for fast detection of failure in laboratory equipment.

The selection and implementation of a quality control (QC) rule is a crucial step in a laboratory in order to ensure the release of correct test results from patients' samples. It is further crucial that failures are identified as soon as possible to prevent incorrect results from being dispatched, as well as to limit the number of samples that have to be reprocessed.

Today's laboratory equipment may offer several different types of diagnostic tests, which can amount to several hundreds of diagnostic tests (each diagnostic test may test a particular substance). In order to guarantee valid test results for a patient's sample, quality control (QC) samples may be used and monitored for each test. It is often a pre-requisite of monitoring quality control samples, to select and implement a statistical quality control rule for each test.

Designing statistical procedures for evaluating laboratory equipment may include a requirement of error detection (such as probability of error detection (Pe _d )) and/or requirements on false rejections, such as a probability of false rejection (P _fr ). If the equipment has a failure resulting in "out-of-control" state of the laboratory equipment, then a high probability of detecting that failure is aimed for (P _ed or "power" of 90% is usually aimed for in the art). However, if the equipment is "in-control", then there should be a low probability that the procedure identifies a failure based on results (P _fr of 5% or 1% is usually aimed for).

Traditionally "Westgard rules" may be used to identify failures in laboratory equipment. Westgard rules may require unnecessary numbers of runs before identifying an out-of-control situation, and may therefore result in unnecessary wait times until results can be deemed reliable, leading to extra costs and longer wait times for patients. Further, Westgard rules provides several alternative rule combinations, without clear selection process, and are therefore complicated to implement and/or require huge training efforts. Westgard rules might be prone to being incorrectly implemented due to the complexity. The Westgard rules are further not adapted to samples including multiple diagnostic tests.

In particular, the deficiencies in the art is solved by a computer-implemented method for determining failures in an equipment, the method comprising the steps:

a) receiving a desired probability of false rejection a desired error detection measure , such as a desired probability of error detection

relating to one or more QC levels (J) of quality control (QC) samples to be processed by the equipment;

b) setting a number of QC runs (R) to one;

c) calculating an error detection measure (E _D ) based at least partially on

R;

d) determining if is below and if so:

increase R by one, and repeat steps c) to d), and

if not:

define a rule for determining failures in the laboratory equipment based, at least partially, on R;

e) receiving or collecting standardized QC results, preferably from at least R runs of QC samples generated by the equipment;

f) applying the rule defined in step d) to the standardized QC results;

g) determining failures in the equipment, if the standardized QC results comply with the rule.

The advantages include a simplified and fast detection of failure in laboratory test equipment. The advantages further include an improved design of statistical rules to quickly identify failures in laboratory equipment with fewer runs. Identifying failures fast improves patient safety by preventing incorrect results to be sent to clinicians, and allow early certainty for time-critical samples. Furthermore, the design is simple and may be automated more easily, facilitating use and increasing safety. The invention is further suitable for equipment performing multiple diagnostic tests.

The received or collected standardized QC results in step e) may be generated from non-standardized QC test results. These QC test results may be measured by the laboratory equipment together with patients' samples. In one embodiment, the rule defined in step d) includes a comparison with a squared distance (d ² ), where d ² is preferably derived from the central chi-square distribution

alternatively from Flotelling's T-squared distribution; and

wherein step f) preferably includes calculating a Mahalanobis distance (M) for the J number of QC levels and the R number of the preferably latest runs, and applying the rule in step f) includes determining if M is equal to or greater than d ² .

The Mahalanobis distance (M) is derived by the equation:

where Z _{j r} is the standardized QC result from QC level j and run r; and wherein Z _jr is preferably given by the equation

where X _jr is QC test results from QC level j and run r, and

m _j is the in-control central tendency, such as mean or median, and s _j is the in- control measure of dispersion, such as standard deviation or interquartile range, relating to QC level j, and where m _j and s _j are preferably derived from equipment validation test results.

The advantages include a single, simple rule that can safely be automated to detect failures in laboratory equipment in few runs.

The non-standardized QC test results may be standardized by subtracting from the non-standardized QC test results, the in-control central tendency/bias (s _j ), and dividing the result of the subtraction by the in-control dispersion/standard deviation (pj) for each QC level j.

The in-control bias and in-control variability may be calculated from validation test results for each QC level. These validation QC results may then be

considered to be in-control (i.e. measured without influence of an out-of-control failure). The validation QC results may be at least 20, or at least 10, initial validation QC results used to derive the in-control bias (e.g. mean of the QC test results) and in-control standard deviation (e.g. standard deviation of the validation QC test results). The validation QC results as well as QC test results may have multivariate normal distribution with low or zero correlations across QC levels and runs.

In one embodiment, relates to a desired probability of systematic error detection and to a critical systematic error ( ^D SEcrit), where ^D SEcrit may be defined A _S Ecru= ^{s ~ z} i-a with z _1-a as quantile of the standard normal distribution that evaluates to l - a and s defined s = (TEa - \b \)/s, where b is bias and s is imprecision aggregated over all QC levels and TEa is the allowable total error; wherein includes a systematic error detection and calculating in

step c) includes deriving from a cumulative distribution function of a non-

central chi-squared distribution H(d ² ; DF, NCP), where NCP = J x R x ^D SEcrit ² ; and wherein step d) includes determining if is below

The advantages include a method that may efficiently identify failure in the laboratory equipment which would result in a shift in test results.

Critical systematic error ( ^D SEcri)t may be defined A _SEcrit = s - z _1-a with z _1-a as quantile of the standard normal distribution that evaluates to 1 - a (the quantile is about 1.65 for a = 5%). The sigma metric a may specify the number of standard deviations being away from the upper limit of the allowable total error TEa when the QC measurement process is in control. The sigma metric may formally be defined s = (TEa - \b \)/s with b as bias and s as standard deviation aggregated over all QC levels.

In one embodiment, E _D relates to a desired probability of random error detection and to a critical random error ( D _REcrit ) , where A _REcrit may be defined by as quantile of the standard normal distribution that evaluates to l - a and a defined a = (TEa - \b \)/s with b as bias and s as imprecision aggregated over all QC levels;

wherein includes a random error detection , and calculating in

step c) includes deriving from a cumulative distribution function of a gamma distribution G(d ² ; SH, SC), where SH = (J x R)/2 and SC = 2 x D _REcrit ² ; and

wherein step d) includes determining if is below

The advantages include a method that may efficiently identify failure in the laboratory equipment which would result in increased imprecision of the test results.

Where relates both to a desired probability of random error detection ( and to a desired probability of systematic error detection determining (in step d)) may include determining if both is below and is below or if at least one of the conditions are met.

In one embodiment, step a) further includes receiving a maximum number of runs (R _Max );

wherein step d) further includes determining if R > R _Max , and if so:

determine that no rule is found; and

where R _Max is preferably 4 for J=2, and where R _Max is preferably 2 for J=3.

In one embodiment, each run is generating data from one, two, or more, preferably less than 100, different diagnostic tests (I);

wherein step g) further includes applying a corrective function to the desired probability of false rejection ; and

wherein the corrective function preferably is

In one embodiment, the steps b) to g) are performed separately for each diagnostic test; and a failure is determined to exist in the equipment if

standardized QC results relating to any of the tests do not comply with a corresponding rule.

The advantages include an increased speed of detecting out-of-control conditions, where several diagnostic tests are measured using QC samples.

In one embodiment, in response to determining failures in step g), the method further includes

h) interrupting a measurement process and/or blocking results, such as results generated after the R or R+l latest QC run.

The advantages include improved patient safety. Where the laboratory equipment utilizes so called bracketed QC strategy, a bracket may be the patient samples between two QC events. The laboratory may hold R number of brackets until the subsequent QC event has passed. If a failure is detected after a QC event, the R number of latest brackets' patient results may be blocked and determined unreliable.

In particular, the object is solved by laboratory equipment for determining failures in the equipment, preferably by implementing the method of any one of the preceding claims, the equipment comprising:

a rule generator unit;

a rule application unit; and

a test result generating unit for generating results relating to at least one QC level (J) and relating to at least one, preferably two or more, different tests (i);

wherein the equipment comprises means for:

a) receiving, by the rule generator unit, a desired probability of false rejection and desired error detection rate such as desired probability of error detection relating to one or more QC levels (J) of quality control (QC) samples;

b) setting, by the rule generator unit, a number of runs (R) to one;

c) calculating, by the rule generator unit, an error detection rate (E _D ) based at least on R;

d) determining, by the rule generator unit, if is below ,

and if so: increase R by one, and repeat steps c) to d), and

if not:

define a rule for determining failures in the laboratory equipment based, at least partially, on R;

e) collecting, by the rule application unit, standardized QC results, preferably derived from at least R runs of QC samples generated by the test result generating unit;

f) applying, by the rule application unit, the rule defined in step d) to the standardized QC results, collected in step e);

g) determining, by the rule application unit, failures in the equipment, if the standardized QC results comply with the rule.

The advantages include a simplified and fast detection of failure in laboratory equipment. The advantages further include an improved design of statistical rules to quickly identify failures in laboratory equipment with fewer runs. In one embodiment, the equipment further comprising a display;

wherein when the rule application unit determines that a failure exists in the equipment, causes the display to indicate that a failure exists in the

equipment.

The advantages include signalling to laboratory personnel that a failure is present in the laboratory equipment resulting in out-of-control patient samples, and that certain amount of patients samples (such as those after the latest R or R+l QC events) are unreliable. The laboratory equipment may also signal a message via wired and/or wireless communication to affected laboratory personnel, clinicians, and/or patients indicating the failure and/or unreliable result and/or estimation of delay for providing reliable results.

In one embodiment, each run is generating data from two or more, preferably less than 100, different diagnostic tests (I); wherein step g) further includes applying a corrective function to the desired probability of false rejection

wherein the corrective func tion preferably includes

wherein the steps b) to g) are performed separately for each diagnostic test; and a failure is determined to exist in the equipment if the standardized QC results relating to any of the tests do not comply with the corresponding rule.

The advantages include an increased speed of detecting out-of-control conditions, where several diagnostic tests are measured in each QC sample.

In one embodiment, the display indicates which of the different diagnostic tests the standardized QC results that do not comply with the rule, relates to.

When a rule is applied separately for each diagnostic test, the laboratory equipment might indicate on the display, which of the diagnostic tests are affected by the out-of-control failure. This assists in locating and correcting the failure.

In one embodiment, the display indicates if the detected failure is a systematic error or a random error. A central tendency (such as mean) of QC results affected by random error failure may not substantially deviate from the in-control central tendency (such as mean), while a central tendency (such as mean) of QC results affected by a systematic failure may result in a central tendency (such as mean) that substantially deviates from the in-control mean.

A dispersion (such as standard deviation) of QC results affected by random error failure may substantially deviate from the in-control dispersion (such as standard deviation). A dispersion (such as standard deviation) of QC results affected by systematic error failure may not substantially deviate from the in-control dispersion (such as standard deviation).

The central tendency and/or dispersion of the out-of-control QC results may therefore be used to determine the type of failure (systematic or random). The indication of the type of error may assist in locating and correcting the failure.

In one embodiment, the equipment, in response to determining failures in step g), further comprises means for:

h) interrupting a measurement process and/or blocking results, such as results generated after the R or R+ l latest QC run.

The advantages include improved patient safety. Where the laboratory equipment utilizes so called bracketed QC strategy, a bracket may be the patient samples between two QC events. The laboratory may hold R number of brackets until the subsequent QC event has passed. If a failure is detected after a QC event, the patient samples of the held R number of latest brackets may be blocked.

In particular, the object is solved by a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of one of the above-stated methods.

The benefits and advantages of the medium is equal, or similar to, the

advantages of the above-mentioned method.

In the following, embodiments of the invention are described with respect to the figures, wherein

Fig. 1 shows steps of a method for determining laboratory test equipment failure; Fig. 2 shows laboratory equipment according to an embodiment of the invention. Fig. 1 shows a method for determining laboratory equipment failure. The method may contain the steps: receiving 110 a desired probability of false rejection

and a desired error detection rate , such as a desired probability of error detection , relating to one or more QC levels (J) of quality control (QC) samples to be processed by the equipment; setting 120 a number of runs (R) to one; calculating 130 an error detection rate based at least partially on R; determining 140 if is below and if so: increase 150 R by one, and repeat

steps c) to d), and if not: define 160 a rule for determining failures in the laboratory equipment based, at least partially, on R; receiving or collecting 170 standardized QC results, preferably from at least R runs of QC samples generated by the equipment; applying 180 the rule defined in step d) to the standardized QC results; and determining 190 failures in the equipment, if the standardized QC results comply with the rule. The method may further optionally include interrupting 195 the measurement process and/or blocking results, such as results generated after the R or R+ l latest QC run.

The step of calculating 130 an error detection measure , may include any, or separately all, of: calculating a systematic error detection from a

cumulative distribution function of a non-central chi-squared distribution H(d ² ;

DF, NCP), where NCP = J x R x ^D SEcrit ² , and calculating a random error detection from a cumulative distribution function of a gamma distribution G(d ² ; SH,

SC), where SH = (J x R)/2 and SC = 2 x D _REcrit ² ·

The step of determining 140 if ED is below ED may, depending on if systematic error detection ) and/or random error detection is targeted, include: determining i is below P _S e _d , and/or if is below If so, increase 150 R by one, and repeat steps 130 to 140, and if not, define 160 a rule for determining failures in the laboratory equipment based, at least partially, on R.

The receiving or collecting step 170, may include standardizing QC test results by the equation Z _jr = (X _jr - m _j )/ s _j , where X _jr is non-standardized (raw) QC test results from QC level j and run r, m _j may be in-control mean, and s _j may be in- control standard deviation relating to QC level j. The runs r = 1 to R, are the R latest runs.

Applying 180 the rule may include determining if a Mahalanobis distance (M) is given by the equation is equal or greater than a squared

distance (d ² ). d ² may be derive from the central chi-square distribution d ² = C ¹ where DF = J x R, or d ² may be derived from Hotelling's T-squared distribution , where DF = J x R.

Fig. 2 shows laboratory equipment 200 according to an embodiment of the invention. The laboratory equipment may include: a rule generator unit 210; a rule application unit 220; and a test result generating unit 230. The laboratory equipment may further optionally include a display 240.

Validation samples, QC samples, and patients' samples may be inserted into the test result generation unit 230 to generate a test results. The rule generator unit 210 may be adapted to receive a requirement on desired probability of false rejection and desired error detection measure and generate a rule as

described above.

The rule application unit 220 may be adapted to receive or generate standardized QC results and apply the rule generated by the rule generator unit 210 as described above to determine if a failure is present in the laboratory equipment. The display 240 may be used to display information to users as described above.