BACKGROUND OF THE INVENTION Spectrophotometers for measuring the composition of a substance by absorption spectroscopy are well known.

For example, oximeters are used to determine concentrations of various hemoglobin components or fractions in blood samples from measuring an absorption spectrum in the visible and/or infrared wavelength range. Such an oximeter is disclosed in EP 210417.

In absorption spectroscopy, determination of a spectrum of a fluid sample is performed by transmission of light through a cuvette containing a part of the sample.

Absorption spectroscopy is based on Lambert-Beer's law according to which the absorbance determined for a sam- ple containing a single optically active component (a dye) is directly proportional to the concentration of the component and the length of the light path through the sample in the cuvette: A () = s () cd (1) in which A (X) is the determined absorbance at wavelength X,

s (X) is the molar extinction coefficient for the compo- nent at wavelength X, c is the molar concentration of the component, and d is the length of the light path through the cuvette holding the sample.

The absorbance A (X) of the sample is defined as the logarithm of the ratio of the light intensity before and after transmission through the sample. In practice the absorbance A (X) is defined as the logarithm of the ratio between the light intensity, Io transmitted through a transparent aqueous reference solution and the light intensity transmitted through the sample: <BR> <BR> <BR> <BR> <BR> A (X) = log io (2) I For samples containing more than one optically active component, the total absorbance Atotal is the sum of the individual components'absorbances since absorbance is an additive quantity. Thus, with Y optically active components in a sample the total absorbance is given by /\ Y/\ Atotal (Ey (X) Cy d y =1 In a sample spectrum, the absorption Atotal (X) recorded <BR> at each wavelength X contains contributions from each of the components in the sample. The magnitude of this contribution and thereby the concentration of each com- ponent in the sample is determined according to

Cy = IX ()) (4) j=1 in which J is the total number of wavelengths Rjat which absorp- tion is determined by the spectrophotometer andKy () is a constant specific for component y at wavelength The vectors Ky (X) may be determined mathematically by using methods such as multivariate analysis, or solving n equations with n unknowns, on data from reference samples.

It is also known to monitor performance of spectropho- tometers, such as oximeters, by a measuring the absorp- tion spectrum of a fluid quality control sample, QC sample, with the spectrophotometer in question.

Known quality control samples specific for blood analy- sis are typically red dye based samples designed to simulate the spectrum of blood. In addition to a red dye, they sometimes contain certain amounts of oxygen, carbon dioxide, and electrolytes at an established pH for determining performance of blood gas and electro- lyte instruments. Synthetic QC samples having an ab- sorption spectrum that closely mimics that of physio- logical blood have not yet been provided.

Quality control of spectrophotometers, such as an ox- imeter, is typically performed by measuring the absorp- tion spectrum of a QC sample comprising three to four different dyes. The dyes are mixed in a proportion so that the QC sample absorption spectrum mimics the ab-

sorption spectrum of blood. A spectrum of a QC sample is measured on the oximeter to be monitored and the pa- rameter values determined by the oximeter are compared with predetermined control limits assigned to the QC sample by a qualified person. If the determined parame- ters are outside the corresponding control limits, servicing of the oximeter is required.

In WO 96/30742 a quality control method for monitoring performance of an oximeter is disclosed. The method comprises measuring the absorption spectrum of a QC sample and comparing it to a standard spectrum of the QC sample. Instrumental errors of the oximeter are con- sidered to be the primary source contributing to the observed difference. Instrumental errors are converted into blood component concentration values so that in- strument errors can be reported in terms understood by the operator of the instrument.

It is an important disadvantage of known quality con- trol methods that, typically, known QC samples comprise 3-4 different dyes, causing long-term stability of the sample to be less than desired. To compensate for this, parameter value acceptance ranges in an oximeter may be widened leading to a more relaxed performance monitor- ing than desired.

It is another important disadvantage of known quality control methods that it is impossible with known qual- ity control methods to distinguish between different types of instrument errors and to determine an individ- ual contribution to deviation in parameter values from a specific type of instrument error. Thus, parameter value acceptance ranges have to be sufficiently wide to accommodate any possible type of instrument error. Fur-

ther, a quality controlled spectrophotometer cannot be diagnosed if the determined parameter values lie out- side the acceptable ranges. For example, a defect spec- trophotometer with a wavelength shift may introduce the same deviation in the determined parameters as seen by dilution of the QC sample.

Future spectrophotometers are expected to facilitate determination of absorption spectra with improved reso- lution whereby instruments of higher precision and specificity are provided. High resolution measurements of spectra makes it more difficult to develop a suit- able QC sample since precision and long term stability requirements are increased.

One of the most significant errors occurring in spec- trophotometers is a wavelength shift. Due to manufac- turing tolerances and drift during use, each spectro- photometer positions a determined spectrum slightly differently along the wavelength axis. Therefore the wavelengths at which absorbance is determined are also positioned slightly differently for different spectro- photometers and thus, determined absorbances will vary for different spectrophotometers.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a quality control method that facilitates the determina- tion of various types of spectrophotometers errors, whereby an accurate diagnosis of an instrument failing the QC test is provided.

An instrument error affects the spectrum of a sample, and specific types of instrument errors affect the spectrum in a distinct way that may be interpreted like the presence of a component in the sample in a differ- ent concentration. For example, a variation of the length d of the light path through the cuvette causes determined absorbances A (X) to vary according to Lam- bert-Beer's law (absorbance is proportional to d), and unintentional dilution of the sample in the cuvette af- fects the determined absorbance in the same way, etc.

It is an important aspect of the present invention that the wavelength shift of a spectrophotometer is deter- mined by forming a Taylor series of a known absorption spectrum or a reference spectrum of a certain component in a sample. After determination of an absorption spec- trum of a sample comprising the component with the known absorption spectrum, the wavelength shift is de- termined.

An absorption spectrum of a sample may be defined by a vector A. (X) comprising at least two elements, each of the elements representing an absorbance of the sample at a specific wavelength j.

A method in quality control of a spectrophotometer is provided, comprising the steps of determining with the spectrophotometer a spectrum A. (k) of a fluid QC sample containing a dye, and <BR> <BR> determining a wavelength shift AR from Ca (X) A (X), in<BR> which CAX (k) is a predetermined coefficient vector pre- viously stored in a memory of the spectrophotometer.

In a preferred embodiment of the method according to the invention the wavelength shift AX is determined af- ter normalisation of the determined spectrum A. (X) with an estimate of the concentration of the dye.

In a further preferred embodiment of the method accord- ing to the invention Ca (X) has been determined from a combination of a reference spectrum Ao (X) of a refer- ence sample containing the dye and a first derivative Ao' (X) of the reference spectrum.

In an approximation, only the first order derivative of the reference spectrum is considered: <BR> <BR> <BR> <BR> <BR> A. (X) = Ao (X) + AX.A(X)(5)<BR> <BR> <BR> <BR> <BR> <BR> <BR> in which Ao (X) is the reference spectrum, Ao' (X) is its<BR> <BR> <BR> <BR> first derivative with respect to the wavelength X, AX is the wavelength shift to be determined, and A. (X) is a spectrum of the sample with the known spectrum A ;, (X) measured by the spectrophotometer in which the wave- length shift is to be determined.

AS may be determined according to various mathematical methods known in the art, e. g. the equation above may be solved for a selected wavelength, the equation may be solved for a set of selected wavelengths and Ak be calculated as an average of the solutions for Ak to the equation, AX may be determined by a least squares fit, Ak may be determined by multivariate analysis, etc.

The invention further provides a method of preparing a spectrophotometer for quality control, comprising the steps of

determining a first reference spectrum Ao (X) of a ref- erence sample containing a dye of a first concentration with a reference spectrophotometer, determining a first derivative Aol (X) of the first ref- erence spectrum of the dye, and determining from at least the first reference spectrum Ao (X) and the first derivative of Ao (X) a mathematical parameter from which a wavelength shift AS of the spec- trophotometer can be determined, and storing the mathematical parameter in a memory of the spectrophotometer.

Further, a spectrophotometer is provided comprising a memory with a mathematical parameter for the determi- <BR> <BR> <BR> nation of a wavelength shift Ak of the spectrophotome- ter, and a processor that is connected to the memory and that is adapted to calculate the wavelength shift AS from the mathematical parameter and from a spectrum A> (X) deter- mined with the spectrophotometer on a fluid QC sample containing a dye.

The mathematical parameter as mentioned above may com- prise the first reference spectrum Ao (X) and the first derivative Ao' (X) of the first reference spectrum Ao (X) at a selected wavelength B0 or at a selected set of <BR> <BR> <BR> wavelengths B0, X Ls etc., or a parameter derived from the spectra, such as the parameter Ca (X).

Preferably, the step of determining a mathematical pa- rameter comprises the steps of calculating a set of calibration vectors Bi (X) accord- ing to <BR> <BR> Bi(#) = siA0(#) + si3A0'(#)(#) = siA0(#) + si3A0'(#) (6) in which i = 1,2,..., N (N>1) and si and si3 are con- stants of selected values, determining a coefficient vector C () constituting the mathematical parameter so that each set of corre- sponding values Si3, Bi satisfies: si3 = c (). B, (.), i = 1,2,..., N (7) <BR> <BR> The step of determining the wavelength shift Ak may<BR> comprise the step of calculating AS from C##(#) # Am(#).

Since the parameter AS is proportional to a total con- centration Cqc of the dye, AS is typically normalised with Cqc or an approximation to cqc/e. g. when the dye is a two-component dye, such as Sulforhodamine B, Ak is preferably normalised with a concentration of a first component of the dye sl. The normalisation of AR with s is desirable when there is a difference between the concentration of the dye in a reference sample from which the reference spectrum was determined, and the concentration of the dye in the QC sample.

Thus, in a preferred embodiment of the spectrophotome- ter according to the invention, the mathematical pa- rameter stored in the memory constitutes a vector CAR from which the wavelength shift AS may be determined.

According to a second important aspect of the inven- tion, the QC sample comprises a dye with two components in a chemical equilibrium where the ratio between the concentration of each component varies with the total concentration of the dye. In this case the shape of the absorption spectrum is dependent on the total concen- tration of the dye. This characteristic of the dye makes it possible to distinguish between a concentra- tion measurement error caused by undesired dilution of the sample in the cuvette, and a measurement error caused by light path changes in the cuvette.

Thus, the method of preparing a spectrophotometer for quality control may comprise determining a first refer- ence spectrum Ao, () of a reference sample containing the dye in a first concentration and determining a sec- ond reference spectrum Ao2 (X) of a reference sample con- taining the dye in a second concentration with the ref- erence spectrophotometer, the dye comprising a first component and a second component in chemical equilib- rium. Mathematically two model spectra A1 (X) and A2 (X) that represent spectral information about the first and the second component, respectively, may be derived from the first and second reference spectra Ao, (X) and Ao2 (X) in such a way that the spectra of the reference samples can be calculated as a weighted sum of A1 (X) and A2 (X).

For example, A1 (X) and A2 (X) may be the individual spec- tra from the two components, respectively, of the dye, or, A1 (X) may be the sum of the individual spectra from the two components while A2 (X) may be the difference between the individual spectra of the two components, etc. Preferably, A1 (X) and A2 (X) are determined from reference spectra Aol (X) and Ao2 (X) by Principal Compo- nents Analysis (PCA).

The spectrum Au, (X) determined by the spectrophotometer is then given by <BR> <BR> Am(#) =(#) = s1A1(#) + s2A2 (#) + ## A0' (#) (8) Each of the parameters sl, s2, and AS may be determined by mathematical methods, such as multivariate analysis on data obtained from reference samples. The step of determining a mathematical parameter may comprise the steps of calculating a set of vectors Bi (X) from Bi (X) = si1 A1 (k) + si2A2 A,()+sA()(9) in which i = 1,2,..., N (N>1) and sil, si2 and si3 are constants of selected values, determining a vector Ca (X) constituting the mathemati- cal parameter so that <BR> <BR> Si3 = C##(#) # Bi(#), i = 1,2,...,N (10) Further, the mathematical parameter may comprise a vec- tor Ci (X) fulfilling that si1 = C1 (#) # Bi(#), i = 1,2,..., N (11) and still further, the mathematical parameter may also comprise a vector C2 (X) fulfilling that si2 = C2 (#) # Bi(#), i = 1,2,..., N (12)

The method in quality control of a spectrophotometer may utilise a QC sample containing the dye in a known concentration Cqc and comprising the first and second components, and may further comprise the steps of calculating parameters s1 and s2 from <BR> <BR> s1 = C1(#) # Am(#) (13)<BR> <BR> s2= C2(#) # Am(#) (14) in which C1 (k) and C2 (X) are the predetermined vectors previously stored in the memory of the spectrophotome- ter, and calculating an estimated concentration cest of the dye from Cest = a Si + b s2 (15) in which a and b are predetermined constants previously stored in the memory of the spectrophotometer, and sl and s2 represents concentrations of a first and a sec- ond component, respectively, of the dye.

Likewise, in a preferred embodiment of the invention the memory of the spectrophotometer may further com- prise vectors C1 (,) and C2 (X) fulfilling the equations (13) and (14).

The memory may also comprise predetermined constants a and b and the processor may be further adapted to cal- culate the concentration Ce, t of the dye according to equation (15)

Cest = a sl + b s2 It is preferred that the dye has a spectrum with a sig- nificant absorbance peak with a steep flank within the measurement range of the spectrophotometer in order to accurately determine small wavelength shifts. For exam- ple, when the sample to be analysed is blood, a wave- length shift of 0.05 nm is sufficient to cause an inac- curate determination of several blood parameters, such as ctHb, s02, FO2Hb, FHHb, FCOHb, FMetHb, etc.

Further, it is preferred that the spectrum of the QC sample resembles spectra of samples, which the spectro- photometer in question is intended to analyse so that performance of the instrument can be monitored.

For example, in blood analysis important blood compo- nents have significant absorbances in the wavelength range from 480 to 670 nm. Thus, a dye with a spectrum resembling a blood spectrum and having a significant absorbance peak in the range from 400 to 800 nm, pref- erably from 480 to 670 nm, and having a steep absorb- ance flank, such as a flank having steepness larger than 40 mAbs/nm, preferably larger than 50 mAbs/nm for a light path length of 100 m, is preferred for use in the methods according to the present invention. The dye should, preferably, also have a molar extinction coef- ficient in the range from 10,000 to 100,000.

The dye may belong to one of several chemical classes, such as cyanine dyes, azacyanine dyes, triarylmethine dyes, acridine dyes, azine dyes, oxazine dyes, thiazine dyes, xanthene dyes, etc. Dyes belonging to the first four classes are typically cationic dyes being water soluble due to the molecule's positive charge. The xan-

thene dyes include the cationic and neutral rhodamines and the anionic sulforhodamines among which Sulforho- damine B is a preferred dye.

According to a preferred embodiment of the invention, the spectrum of reference samples containing the dye in at least two different concentrations is determined, e. g. by an accurate reference instrument of the same type as the spectrophotometer to be quality controlled, at a selected set of wavelengths. Then the coefficient vectors C1 (X), C2 (X) and the constants a and b are determined, e. g. by multivariate analysis, and stored at the time of manufacture in the memory of the spectrophotometers to be quality controlled by fluid QC samples when put into their normal use.

On manufacture of a QC sample the concentration cqc, the ratio s2/sl denoted Qref and an initial wavelength shift Akqc may be determined by a reference spectropho- tometer. The initial wavelength shift of the QC sample emerges mainly from a variation in the composition of the solvent of the dye in the QC sample.

A label, such as a bar-code label, a magnetic label, etc, may be attached to each of the QC samples contain- ing one or more of the values cqc, Qref and qc in ques- tion. Alternatively one or more of the values may be printed in a bar code on a paper sheet following a set of QC samples. The values appearing on the labels or paper sheet are designated assigned values.

During quality control of a specific spectrophotometer, the assigned values of c, Qref and Akqc are read by the spectrophotometer and the values are stored in its mem- ory. Then the spectrum of the QC sample is determined

and sl, s2, and AS are determined as previously de- scribed. The determined values for Qest = szsl 0 and Cest are also calculated and compared to the assigned values of Qref, Akqc and cqc respectively.

A possible dilution of the QC sample may be determined <BR> <BR> <BR> <BR> from a difference between Qest and Qref, and the combined effect of dilution and deviations in length d of the light path through the cuvette may be determined from a difference between cest and c.

The estimated parameter values, such as AR, ce, ; t, and may may be used for determination of parameter values of samples, the analysis of which the spectrophotometer is intended for, so that the outcome of the quality control procedure can be reported by the instrument in quantities meaningful for an operator of the instru- ment.

For example, in an oximeter for determination of blood parameter values, the theoretical modifications to one or several predetermined standard blood spectra caused by a measurement error corresponding to one of the pa- rameters AR, cest, and Qest determined in the quality control procedure may be calculated by the oximeter.

From the modified spectra, the oximeter may calculate corresponding blood parameter values to be reported to the operator of the instrument.

The predetermined standard blood spectra may either be stored in the memory of the oximeter, or they may be derived mathematically by the processor in the oximeter from predetermined spectra of each blood component com- prised in the standard blood samples.

In a preferred embodiment of the invention predeter- mined control limits for the reported blood parameter values are printed on a sheet of paper following a set of QC samples. The operator may compare blood parameter values reported by the oximeter with the predetermined control limits on the paper sheet, and determine whether the reported values are within the control lim- its.

The predetermined control limits may also be stored in a label of the QC sample which label is read by the ox- imeter so that the oximeter is adapted to perform the comparison between the reported blood parameter values and the corresponding control limits.

According to a third important aspect of the invention, a method for repressing absorption spectra of interfer- ing components or substances in a fluid sample, is also provided.

In the present context an interfering component in a sample is a component other than the preselected compo- nents for which the spectrophotometer is adapted to re- port parameter values, and the presence of said inter- fering component in the sample may interfere with the absorption spectrum of at least one of said preselected components.

In a determined sample spectrum, the absorbance A. (k) recorded at each wavelength X contains contributions from each of the components in the sample including said interfering components. The magnitude of the con- tribution and thereby the concentration of each compo- nent in the sample is determined according to equation (16) or equation (17) below

or the equivalent form <BR> <BR> Cy=Ky()-A,()(17) The vectors Ky (X) may be determined mathematically by using methods, such as multivariate data analysis, or solving n equations with n unknowns from data obtained from reference samples. By including one or several in- terfering components or substances in the reference sample, of which the reference spectrum is determined, one or several of the vectors Ky (X) corresponding to one or several of the interfering components may be de- termined. The vector or vectors Ky (X) corresponding to the interfering components are generally designated Kint (X) and stored in the memory of the spectrophoto- meter together with the vectors Ky (X) * The spectrophotometer may further provide one or several predetermined vectors, Aint (X), representing spectral information of the interfering components.

Each Aint (X) is obtained at a reference concentration Cref, whereby the spectrum of any interfering component may be derived at the determined concentration of the component according to Lambert-Beer's law, equation (1).

In an embodiment of the invention, the effect of the interfering components on determined blood parameter values is minimised by following a three stage process, in the following denoted"repression of spectra of in- terfering components".

First stage is to determine the concentration of inter- fering components in the sample. Second stage is to de- termine a modified spectrum of the sample by subtract- ing the spectrum of the interfering component of the determined concentration from the measured spectrum Am(#) of the sample. Third stage is to determine con- centrations of blood components cy and parameter values of blood components from the modified spectrum.

According to the invention, a spectrophotometer with repression of spectra of interfering components in a fluid sample is provided, for determination of a con- centration cy of a component y of a sample and wherein the memory further comprises at least one vector Aint (X) representing spectral infor- mation of an interfering component in the sample at a concentration cint, and at least one vector Kint and wherein the processor is further adapted to calculate the concentration cint of the interfering com- ponent according to <BR> <BR> <BR> <BR> <BR> Cint = Kint (X) Am (X) (18) and if cint is greater than a predetermined threshold value, Cref modify the measured spectrum Amod (X) accord- ing to <BR> <BR> <BR> <BR> <BR> AM=A,()-A,)(19) Cref Amod (X) being the modified spectrum, and

determine cy from the modified spectrum Amod (X) accord- ing to <BR> <BR> <BR> <BR> <BR> c=K).A,,()(20) whereby the effect of interfering components on deter- mined concentrations cy is minimised.

The measured spectrum is only modified if the deter- mined concentration of the interfering component is above a predetermined threshold value. This is because the modification of the measured spectrum creates some undesired"process noise"in the modified spectrum, due to an uncertainty in the estimate of the spectrum of the interfering component. This addition of"process noise"in the modified spectrum is only justified when the concentration of the interfering component in the sample is larger than the threshold value.

An oximeter for blood analysis may provide several pre- determined vectors for interfering components or sub- stances of clinical importance and provide correspond- ing values of the vectors Ki. t (),) in the memory. The in- terfering components may be chosen among components, which have previously caused significant interference in oximetry measurements, such as Fetal Hemoglobin, Bilirubin, Cardio Green, Evans Blue, Methylene Blue, Intralipid, HiCN, SHb, etc. By repressing the spectra of these components an oximeter with better precision in measurement of blood parameter values than currently available instruments is provided.

BRIEF DESCRIPTION OF THE DRAWINGS The invention will now be described with reference to the drawings, wherein Fig. 1 is a block diagram of an oximeter according to the invention, Fig. 2 is a schematic diagram of a wet section of an oximeter according to the invention, Fig. 3 shows main components of a spectrometer, i. e. the optical part of an oximeter according to the inven- tion, Fig. 4 shows compositions of QC samples levels 1-4.

Fig. 5 shows absorption spectra of Sulforhodamine B in three concentrations, Fig. 6 shows two normalised model spectra determined with Principal Component Analysis from Sulforhodamine B, Fig. 7 is a table comprising parameter values of blood samples each related to one of QC sample levels 1-4, Fig. 8 shows absorption spectra of four standard blood samples related to quality control levels 1-4, Fig. 9 is a graph of a variable Fneon plotted against the wavelength of light striking two photodiodes in the spectrometer, Fig. 10 shows response curves of photodiodes located in wavelength channels 70,71 and 72.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT Fig. 1 is a block diagram comprising a spectrometer 1 in an oximetry module (not shown) connected to a printed circuit board 2 with a data cable 6 comprising electrical conductors. The printed circuit board 2 con- trols and collects data from the spectrometer 1. The

data collected are transmitted to a data processing unit 3 comprising a memory (not shown) and a processor (not shown). Values of predetermined coefficient vec- tors Cl (X), C2 (X) and Ca (X) are stored in the memory. A barcode reader 5 is adapted to read data from bar-code labels mounted on QC samples or on a paper sheet en- closed with a set of samples, and transmits data to the data processing unit 3 via a data management computer 7. A power supply module 4 supplies power to the oxime- try module from a mains connection.

Fig. 2 is a schematic diagram of a wet section of an oximeter according to the invention, wherein a blood sample (not shown) is entered into the oximeter through an inlet probe 20. The sample is transferred to a cu- vette 74. A preheater 15 is positioned along the sample path 30 to heat the sample to a substantially constant temperature of 37 °C. Pumps 10 are used to pump liquids and gasses through the wet section.

Fig. 3 shows the main components of the spectrometer 1, wherein a light beam 75 with constant intensity is transmitted from a halogen lamp 70 to the cuvette 74 which comprises the blood or QC sample and is included in a hemolyzer 79. The blood sample is hemolyzed by means of ultrasonic waves. Hemolyzing is a process, which ruptures the walls of the red blood cells in the sample, thereby making the blood cells release their content of hemoglobin. The light beam 75 is transmitted to the cuvette 74 through an infrared filter 71, and a biconvex lens 72. After passing through the cuvette 74, the light beam 75 is transmitted to a measurement sec- tion 76, by means of an optical fibre 77. The light beam 75 passes through a thin slit 78, whereby the beam

75 is directed towards a concave grating unit 80, dif- fracting the light beam 75 according to wavelength.

The concave grating unit 80 focuses light on a photodi- ode array 83, to which a diffracted light beam 82 is transmitted. The photodiode array 83 may consist of 128 photodiodes, and the array 83 is arranged in such a manner that light comprising a range of wavelengths of approximately 1.5 nm in the diffracted light beam 82, strikes a photodiode (not shown), which converts the light into a current substantially proportional to the light intensity which strikes it. By measuring the value of the current in each of the 128 photodiodes of the photodiode array 83, a discrete intensity spectrum of the light beam 82 after transmission through the sample is produced. From this intensity spectrum an ab- sorption spectrum of the blood sample comprised in the cuvette 74 may be determined by the oximeter.

The absorption spectrum is measured in 128 channels lo- cated in the wavelength range 478-672 nm in the pre- ferred embodiment of the invention. A channel is, in the present context, the part of the spectrum which is transmitted to a particular photodiode in the diode ar- ray 83.

According to the invention a wavelength shift of the oximeter is determined in the quality control proce- dure. It is preferred that four different types of quality control samples (QC samples levels 1-4) are provided, cf. Fig. 4. The QC levels comprise Sulforho- damine B in different concentrations. Increased reli- ability in the quality control of the oximeter is pro- vided by measuring the absorption spectrum of QC sam- ples at several QC levels. By utilising QC samples com-

prising Sulforhodamine B in different concentrations, it is ensured that the oximeter measures blood parame- ters correctly over a wide range of component concen- trations in blood samples.

In solution Sulforhodamine B shows long term stability.

The steep absorbance flank allows an accurate determi- nation of the wavelength shift of the oximeter, since even very small wavelength shifts produce a large change in the measured absorbance at a given wavelength of a Sulforhodamine B containing sample.

In aqueous solution Sulforhodamine B is a dye with two components in a chemical equilibrium where the ratio between the concentration of each component in the dye varies with the total concentration of the dye. In this case the shape of the absorption spectrum is dependent on the total concentration ci of the dye. This may be seen in Fig. 5, which shows three absorption spectra Aol (X) 110, Ao2 (X) 111 and Ao3 (X) 112 of Sulforhodamine B samples determined at the total concentrations 2.5058 mmol/kg, 1.6705 mmol/kg and 1.0023 mmol/kg, respec- tively. The Sulforhodamine B samples correspond to QC levels 1-3 as shown in Fig. 4.

Mathematically, two model spectra A, (X) 105 and A2 (X) 106 as shown in Fig. 6 may be derived from at least two reference spectra, e. g. Aol (X) 110 and Ao2 (X) 111 of Fig. 5, wherein the two model spectra represent spec- tral information about a first and a second component, respectively of Sulforhodamine B, in such a way that the spectrum of the dye can be calculated as a weighted sum of A, (X) and A2 (k).

The two model spectra are, preferably, determined by Principal Component Analysis (PCA), whereby two or- thogonal vectors are determined constituting the mathe- matical model spectra, Ai (,) and A2 (X). A set of scores or parameters si1and si2 is also provided by the PCA analysis for each concentration of the dye, as the spectrum of the dye at a concentration ci can be calcu- lated as a weighted sum of model spectra A, (X) and A2 (X) and their corresponding scores or weights siland si2.

The PCA analysis may be provided by several computer programs, which are commercially available. The program used in the present embodiment is the"Unscrambler".

The two model spectra A, (, %) 105 and A2 (X) 106 shown in Fig. 6 are determined by PCA from the three reference spectra Aol (X) l Ao2 (X) and Ao3 (X) with"Unscrambler".

The reference concentrations of the dye in the solution at which the reference absorption spectra Aol (), A () and Ao3 (X) are measured, are determined from the weight of the dye, Sulforhodamine B in powder form and the volume of the solvent. The reference absorption spectra are determined by measuring the absorption spectra of 5 samples containing Sulforhodamine B at each reference concentration, and determining an average value for the reference spectrum for each concentration. The refer- ence absorption spectra of the samples are measured by a reference oximeter, which by definition has a zero wavelength shift.

In practice, an oximeter not specifically appointed and handled as a reference oximeter will always exhibit some wavelength shift AR whereby a measured absorption spec- trum Am (X) of a sample will differ slightly from the reference spectrum Ao (/) of the same sample measured on

the reference oximeter. The relationship between the measured spectrum A. (X) and a reference spectrum Ao (X) and the model spectra is for small wavelength shifts according to equation (8) <BR> <BR> Am(#) =(#) = s1A1(#) + s2A2 (#) + ## AS A0'(#)<BR> <BR> wherein Ak Ao' (X) is the first term in a Taylor series of the reference spectrum Ao (k).

The first derivative of the reference spectrum Aol (X) is preferably calculated in approximation as a first de- rivative of the model spectrum A,' (k). The approximation is justified since the values of the scores sil for the model spectra A1 (X) are found to be much higher than the values of the scores si2 for the model spectra A2 (k), of Sulforhodamine B in relevant concentrations ci, so that <BR> <BR> A.'()=s'()+s,A,'(.)SiA\()(21) whereby the measured spectrum Ami (X) may be approximated by Ami(#) = si1A1(#) + si2A2(#) + ##isi1A1'(#)(22) Akisi1, si1, si2 are the scores or the constants corre- sponding to a concentration ci. <BR> <BR> <P>Coefficient vectors C1 (#), C2(#) and CAX (.) are, prefera- bly, determined by multivariate analysis from the scores and the corresponding determined absorption spectra.

The multivariate analysis starts by generating a table with 64 rows and 4 columns. The first three columns in this table comprise selected values of either one of the

scores As, s, s, and the last column comprises the corresponding calculated value of the spectrum Ami(#).

Each row constitutes a calibration vector, and the en- tire table constitutes 64 calibration vectors.

The 64 values of each score appearing in one and the same column are evenly distributed between: wherein A2 (Xj) denotes the summation of squared absorban- ces across 128 wavelengths of the particular spectrum that corresponds to a particular score; i. e. the values of the score sil are evenly distributed between 0 and reciprocal of (square root (Al2 (X)).

The next step in the multivariate analysis comprises to determine from the table the coefficient vector Cl (X) by Principal Component Regression so that each set of scores sil, and the corresponding spectrum Ami(#), satis- fies <BR> <BR> si1 = C1 (#) # Ami(#) (23) From the table the coefficient vector C2 (X) is deter- mined by Principal Component Regression so that each set of scores si2 and the corresponding spectrum Ami(#), satisfies si2 = C2(#) # Ami(#)(24) From the table the coefficient vector C (X) is deter- mined by Principal Component Regression (PCR) so that each set of scores ARisiland the corresponding spectra Ami (X), satisfies

##i si1 = C##(#) # Ami(#) (25) Further, it is assumed that the following relation between the calculated scores and a total concentration, ci of the dye exists ci = a sil + b si2 (26) wherein constants a and b may be found by several meth- ods, preferably, by inserting the determined scores from the total concentrations, ci of the dye of concentra- tions 2.5058 mmol/kg and 1.0023 mmol/kg in equation (26) and solve the resulting two equations with two un- known quantities, for a and b. The determined values of a, b are: a=0.1425; b=0.0931, so that equation (26) is determined as ci = 0.1425 sil + 0.0931 si2 (27) The validity of equation (27) may be checked by insert- ing scores sil, si2 from reference solutions with total concentrations ci of Sulforhodamine B not used in the determination of a and b. Thereby, the validity of equation (27) has been confirmed experimentally.

In field use of the spectrophotometer the coefficient vectors are applied as follows: From the coefficient vector, Cl (X) a score or parameter value, si may be determined according to equation (13) s, = Ci (X) Am (X)

wherein Am (X) is a measured spectrum of a QC/reference sample.

From the coefficient vector, C2 (X) a score or parameter value, s2 may be determined according to equation (14) S2 = C2 ().A() wherein A. (X) is a measured spectrum of a QC/reference sample. <BR> <BR> <BR> <BR> <BR> <P>From the coefficient vector CAX (X) a score or parameter value ARsl, which is proportional to the wavelength shift may be determined according to <BR> <BR> <BR> <BR> <BR> AS sl = CA) Am (X) (28) wherein Am (X) is a QC/reference sample.

Determined sl, s2 scores may be interpreted as the equivalent concentrations of the first and the second component of the dye, respectively. The first component corresponds to the mathematical model spectrum A, (k), and the second component corresponds to the mathemati- cal model spectrum A2 (X).

The determined coefficient vectors Ca (X), Cl (X) and C2 (X) are stored in a matrix in the memory of the ox- imeter at the time of manufacture. The determined con- stants a, b are also stored in the memory of the oxime- ter at the time of manufacture.

QC samples are, preferably, manufactured in lots, which may comprise 40,000-50,000 samples. The lot values of cqc, Qref and ##qc are, preferably, determined during

manufacturing by measuring 5-10 samples on 3 reference oximeters. The oximeters have been adjusted to report exact parameter values on a standard blood sample.

Average values of each of the measured parameters Cqc Qref and A are calculated and preferably stored on a bar-code label attached to each of the QC samples.

During a quality control procedure of an oximeter in normal operation, e. g. at a hospital, the values of cqc, Qref and ARqc are read from the bar-code label of the QC sample by a bar-code reader and stored in the memory of the oximeter.

Then the absorption spectrum of the QC sample is deter- mined. An estimated concentration of Sulforhodamine B in the QC sample may be determined by the measured ab- sorption spectrum Au (k) by equation (26) as Cest = 0.1425 Si + 0.0931 s2 since the values of sland s2can be determined by the measured absorption spectrum Au (X) and the vectors Cl (X) and C2 (X) according to equations (13) and (14). The ra- tio between si and s2 is determined and denoted Qest- An estimate of a score proportional to the wavelength shift of the oximeter is provided by equation (25) AX S1 = C (). A ().

Since the value of s1has been determined, the value of the wavelength shift of the oximeter is determined by dividing the score Ak sl with sl

=M-) (29) si The length of the cuvette light path do in the oximeter is, preferably, determined by measuring an absorption spectrum Am (X) of a Sulforhodamine B reference solu- tion. The concentration of Sulforhodamine B, cref, is, preferably, provided as an assigned value.

To determine the value of do, the absorption spectrum A. (,) of the reference solution is measured, and an es- timate of the concentration c, st of the dye is calcu- lated by the processor in the oximeter according to equations (26), (13), (14) by utilising predetermined coefficient vectors Ca (X), Cl (X) and C2 (X) and con- stants a, b stored in the memory of the oximeter as previously described.

The concentration cest of the reference solution deter- mined by the oximeter is utilised to calculate an ac- tual value of the cuvette light path length, do, in the oximeter according to <BR> <BR> <BR> <BR> <BR> <BR> <BR> d. =d-(30)<BR> <BR> <BR> <BR> C ref wherein dref is a reference value of the cuvette light path length, which is previously stored in the memory of the oximeter. The calculated value of do is subse- quently stored in the memory of the oximeter.

The difference between the value of AS determined for the Sulforhodamine B reference solution and the as- <BR> <BR> <BR> signed value Agj ; for the reference solution is util-

ised to shift the subsequently measured spectra along the wavelength axis.

The absorbance A (X) of a fluid sample is measured by the oximeter by determining the logarithm of a light intensity Iotransmitted through a transparent aqueous reference solution divided by the light intensity I transmitted through the fluid sample in question, ac- cording to equation (2) Along 0 I Io is denoted the zero point intensity, and is measured automatically at every calibration of the oximeter with said reference solution.

During a quality control of the oximeter, a determined value of Ce. t may be compared with the corresponding value Cqc read from the label of the QC sample. A dif- ference between the values may originate from two of the variables in Lambert-Beer's law, equation (1) A (X) = £ (X) c d.

Tt applies that either the cuvette light path length d in the oximeter is different from the do value stored in the memory of the oximeter, which causes a higher or a lower value of the measured absorbance, or the meas- ured concentration Cest of the dye deviates from the value of cqc.

The determined concentration Ce, t may deviate from the value of Cqc due to errors in the wet section of the oximeter, such as defect tubes, defect pumps, errors in the cuvette, etc. It may all lead to undesired dilution

of the sample. However ces mamy also be different from cqc due to an incorrect light path length do of the cu- vette.

If there is a difference between c,,,, and cqc, and the value of Qref being equal to Qestl the difference between the estimated concentration and the reference concen- tration values may be caused by a difference between the light path length d,, of the cuvette as calculated during calibration and the reference value drefof the length determined during manufacture.

If there is a difference between Ce, ; t and cqc the value of Qref being different from Qest'the sample may be di- luted. A dilution causes the concentration of the dye to be smaller than cref and further causes a shift in the chemical equilibrium between the components sl and s2 which causes the value of Qest to deviate from Qref The determined differences between measured parameters and Qest and the corresponding parameters read from the bar-code label of the QC sample may be re- ported by the oximeter to the operator e. g. by means of a printer. A printed message may comprise information as to which of the measured parameters caused the QC sample to fail the quality control. Together with a printout of the failing parameter a message suggesting which part of the oximeter needs repair or service, may be included. For example, the printed message may rec- ommend a repair of the measurement section 76 of the spectrometer 1, if the measured wavelength shift Ak is larger than a predetermined threshold value stored in the memory of the oximeter.

In a preferred embodiment of the invention the measured parameters of the QC sample are used to modify spectra of standard blood samples corresponding to either of the QC levels 1-4.

In Fig. 7 the figures in columns 2-7 of each row define a standard blood sample composition, and column 1 shows the related QC level. For each of the four standard blood samples a corresponding standard blood spectrum as shown in Fig. 8 may be derived mathematically by the processor in the oximeter from predetermined spectra of each blood component comprised in the standard blood samples. The predetermined spectra of each blood compo- nent are, preferably, stored in the memory of the ox- imeter during manufacture.

Each blood component parameter value in the table in Fig. 7 has an attached plus/minus limit value. The limit values are calculated errors, which would be pro- duced by a measurement of parameter values in the stan- dard blood sample with an oximeter having a wavelength shift of plus and minus 0.05 nm, respectively, as the only measurement error. For example, the value of blood component FCOHb in a level 1 sample would be measured to 5.34 % or 6.66 % instead of the correct value of 6.00 %. Thus, even very small wavelength shifts in the oximeter, introduces significant errors in the measured blood parameter values, thereby illustrating the impor- tance of quality controlling the oximeter for wave- length shifts.

By determining the modifications to the mathematically derived standard blood spectrum related to the level of the actual QC sample under test resulting from the pa- rameters AS, Cest and optionally also Q, st, determined in

the QC procedure, the oximeter may use the modified spectrum to calculate corresponding blood parameter values. The parameter values are reported to the opera- tor of the oximeter, and the operator may compare them with assigned control limits for the actual QC level.

The effect of the instument errors revealed in the QC procedure on values reported for a blood sample with unknown blood parameter values may, e. g., appear from the deviations between the reported parameter values and the values of the relevant standard blood sample of Fig. 7.

Fig. 8 shows absorption spectra for each of standard blood sample, which absorption spectra are used in the oximeter for quality control levels 1-4. The spectra corresponding to levels 1-4 are 120,121,122,123, re- spectively. Each spectrum has a corresponding cref value corresponding to a Sulforhodamine B concentration.

The above modification to the standard blood spectra shown in Fig. 8 resulting from the parameter Ak is a shift along the wavelength axis corresponding to the difference between AR and Akqc 1 Akqc being either an as- signed value or a predetermined fixed value stored in the memory of the oximeter. The modification of the standard blood spectra resulting from the parameter cest is a modification of the individual absorbances with the ratio Cest/Cref * By adopting this method of converting determined meas- urement errors introduced by the oximeter into parame- ter values of blood samples, instrument errors are re- ported in terms which are easily understood by the op- erator of the oximeter.

By noting which of the blood parameters failed the con- trol, it may be possible to determine which of the measured parameters AR, c,,,, and Qest caused the quality control to fail, and thereby determine which part of the oximeter that needs repair or service.

The relation between blood parameters that failed the quality control by being outside their corresponding control limits and the measured values of parameters and Qest and thereby an error diagnosis of the oximeter may, preferably, be comprised in a service manual for a repair technician.

According to the invention a method is provided for re- pressing absorption spectra of one or several interfer- ing components or substances contained in a blood sam- ple in the oximeter. Preferably, the oximeter is adapted to repress the spectrum of Fetal Hemoglobin, which is known to cause significant interference in ox- imetry measurements.

In a determined blood sample spectrum, the absorbance recorded at each wavelength X contains contribu- tions from each component in the sample. Interfering components are naturally treated as the other compo- nents. The magnitude of the contribution and thereby the concentration of each component in the sample is determined according to equation (17) Cy = Ky (X A).

The vectors K (X) are predetermined and stored in the memory of the spectrophotometer.

By including a Fetal Hemoglobin component in a blood sample, of which the reference spectrum is to be deter- mined, a vector Kfetal (X) corresponding to the concentra- tion of Fetal Hemoglobin in the sample, is determined.

Preferably, the oximeter further provides a predeter- mined vector A (), representing the difference spectrum between Adult Hemoglobin and Fetal Hemoglobin.

The vector Afetal (X) is, preferably, determined at a ref- erence concentration cfetalof 1 mmol/L.

The effect on determined blood parameter values due to the presence of Fetal Hemoglobin in the blood sample, is minimised by repressing the spectrum of Fetal Hemo- globin.

The first stage in the repression process comprises the determination of the concentration of Fetal Hemoglobin in the blood sample, according to equation (18) Cfe.al=Kfetal()'A).

The second stage comprises the determination of a modi- fied spectrum by subtracting the difference spectrum at the determined concentration from the measured spectrum Am(#) of the blood sample, if cfetal is greater than a predetermined threshold value, according to equation (19) Amod(#) = Am(#) - Cfetal/1 Afetal(#) wherein Amod (X) is the modified spectrum and Cref = 1 mmol/L.

If cfetalis smaller than the predetermined threshold value the modified spectrum is set equal to the meas- ured spectrum Au (k).

The third stage comprises the determination of concen- trations of blood components cy from the modified spec- trum Amod (X) l whereby the effect of Fetal Hemoglobin in the blood sample on determined concentrations cy of blood components is minimised.

By repressing the spectrum of Fetal Hemoglobin auto- matically, an oximeter is provided with an increased precision in measured blood parameter values, and an easier operation than currently available instruments.

Fig. 9 is a graph of a variable Fneon plotted against the wavelength of light striking two photodiodes in wavelength channels 70 and 71 of the photodiode array 83 in the spectrometer 1 shown in Fig. 3. The spectro- meter 1 comprises a neon glow lamp (not shown), which emits at least one spectral line at a highly accurate reference wavelength of 585.25 nm, suitably positioned within the preferred wavelength range from 480 to 670 nm. The accurate wavelength of the emitted spectral line is used in the oximeter as a reference wavelength against which the location of the 128 wavelength chan- nels of the array 83 is adjusted. To utilise the refer- ence wavelength a variable Fneon is defined as Fneon = R (70)/R (71) (31) wherein R (70) and R (71) are the magnitudes of the cur- rent or the response in each of the photodiodes located in channels 70 and 71. Fneon is also equal to the ratio between the light intensity striking photodiodes in

channels 70 and 71, due to the linear relationship be- tween the current in a photodiode and the light inten- sity which strikes it. For example, if Fneon = 1 the light intensity striking diode 70 is equal to the light intensity striking diode 71, which means that the ref- erence wavelength is positioned exactly between the channels 70 and 71. Fneon is used as a variable that de- fines the position of the light of the reference wave- length emitted from the neon lamp relative to the wave- length channels in the spectrometer 1. This character- istic of Fneon is utilised during field operation of the oximeter, where the value of Fneon is measured at prede- termined time intervals, and compared with a reference value Focal stored in the memory of the oximeter during manufacture.

The spectrometer 1 is scanned with light emitted from a high precision monochromator in the wavelength range 585.25 +/-7.5 nm during manufacture. A response curve for the photodiode located in channel 71 is measured.

An example of a measured response curve is 131 shown in Fig. 10. A calibration algorithm comprised in the mem- ory of the oximeter calculates a corresponding response curve for channel 70 by shifting the wavelength axis.

The calibration algorithm further calculates a wave- length calibration table comprising values of the vari- able Fneon and the corresponding value of the wavelength of light emitted from the monochromator by using the determined response curves of channels 70 and 71. The oximeter stores determined values of the wavelength calibration table in the memory. A reference value of is determined during manufacture by activating the neon lamp and measuring the response of channels 70 and 71, as previously described. The refer-

ence value of Focal is stored in the memory of the oxime- ter.

The data comprised in the wavelength calibration table may be displayed graphically as shown in Fig. 9.

A calibration program measures the current temperature of the spectrometer 1 between two blood sample measure- ments in normal operation of the oximeter. The cuvette is always cleaned with a transparent rinse solution be- tween two blood sample measurements. The current meas- ured temperature of the spectrometer 1 is compared with a previous temperature measurement which was performed at the time of the previous neon lamp activation and stored in the memory of the oximeter. The calibration program determines whether the current temperature value deviates more than 0.3 °C from the previous tem- perature value, and performs a measurement of the cur- rent value of Fneon if this is the case.

The graph in Fig. 9 is now used to illustrate how a wavelength shift of the oximeter is determined and com- pensated during a period of time between two blood sam- ple measurements, wherein the cuvette is rinsed. A first value of Fneon denoted Focal corresponding to a first value of the wavelength denoted cal are shown in the graph, and the value of Focal is determined, as previ- ously described. A second value of the variable Fneon denoted Fact may be measured by the oximeter between two blood sample measurements. By utilising the predeter- mined wavelength calibration table comprised in the memory of the oximeter a second value of wavelength<BR> denoted k, ct corresponding to Facs mamy be determined. The<BR> value of k, ct may be determined from the discrete values<BR> of the variable X comprised in the calibration table

according to well-known mathematical interpolation methods such as linear interpolation, polynomial inter- polation, cubic spline interpolation, etc.

A wavelength shift AS of the spectrometer may be deter- mined from the difference between the determined value <BR> 'act and the calibration value The determined wave- length shift Ak of the spectrometer 1 may be utilised to compensate a measured absorption spectrum A. (X) of a fluid sample by determining a modified absorption spec- trum Amodi (X) of the sample, wherein the effect of the determined wavelength shift AS on absorbances in the measured spectrum Am (X) is removed.

The modified spectrum is, preferably, determined by first utilising a cubic spline function to generate in- terpolated absorbance values between the discrete val- ues at the 128 wavelengths in the measured spectrum Au (,). The modified spectrum A.. di (k) is determined by shifting the wavelength of each measured absorbance value in Am (X) sequentially with an amount equal to Ak and determine a corresponding interpolated absorbance value for the modified spectrum.

The provision of a spectral lamp, preferably a neon lamp, having at least one spectral line within a de- sired wavelength range enables the oximeter to perform highly accurate measurements of the wavelengths of light absorbed by a sample by comparing the determined wavelength of said at least one spectral line with the assigned wavelength of the spectral line stored in the memory of the oximeter, calculating the possible wave- length shift, and compensating the determined absorb- ance of the sample for said wavelength shift. Accord- ingly, the determined absorption spectrum by the spec-

trometer 1 is being compensated for wavelength shifts resulting from manufacturing tolerances and temperature drift during the use of the oximeter, thereby providing accurate measurements of blood parameter values.

Fig. 10 shows three response curves 130,131 and 132 of photodiodes located in the corresponding wavelength channels 70,71 and 72. The x-axis of the graph is the wavelength in nm of the light striking the diodes, and the y-axis of the graph is counts. The wavelength dis- tance between the peak points of e. g. response curve 130,131 is approximately 1.5 nm, which is the channel distance between all the 128 adjacent wavelength channels of the diode array 13.