Method and system for predicting and evaluation of late onset sepsis in neonatal infants

Abstract

Provided herein is a system for dynamic assessment of a prediction of late onset sepsis, LOS, in a test neonatal subject, a test subject, the system comprising a processor configured to perform steps of a method comprising: - receiving a parameter time series (120,a to i) of a measurement of at least one parameter, a test parameter, measured of the test subject after birth of the test subject during a test session; - determining a prediction time series (130a to g) of sequential prediction scores for LOS, TLOSPS(ptd), of the test subject, wherein - each TLOSPS(ptd) within the prediction time series is determined at a determination time point, ptd; - each TLOSPS(ptd) (e.g. 130,a or 130,e) is determined from the at least one parameter time series (e.g. 120,a or 120,a) limited to a parameter time window, PTW, - the PTW spans up to the ptd, and extends back in time according to a size of the PTW; - dynamically evaluating the TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd).

Description

[0001] METHOD AND SYSTEM FOR PREDICTING AND EVALUATION OF LATE ONSET SEPSIS IN NEONATAL INFANTS

[0002] Field of the invention

[0003] The invention is broadly in the field of prediction of late onset sepsis in neonates, and in particular relates to an evaluation of that prediction.

[0004] Background to the invention

[0005] Late-onset sepsis in neonatal infants is a serious problem in neonatal intensive care units. Diagnosis can be difficult because clinical symptoms are not specific, and no reliable biochemical marker has been identified. The gold standard for detecting LOS is blood culturing, in which a blood sample is removed from the subject and a culture is grown therefrom, however, it takes between 12 and 48 hours to detect microbial growth from the moment of sample extraction. A microbial analysis is typically ordered after a suspicion of late- onset neonatal sepsis (e.g. caused by a rising temperature), however, the 12 to 48 hours delay in receiving the result leaves clinicians with a difficult choice of pre-emptively treating the subject with antibiotics so leading to unnecessary antibiotic use if the subject is LOS negative, or delaying treatment which leads to a worsening condition if the subject is LOS positive. A method for predicting late-onset sepsis in neonatal infants is needed.

[0006] Summary of the invention

[0007] Provided herein is a system for dynamic assessment of a prediction of late onset sepsis, LOS, in a test neonatal subject, a test subject, the system comprising a processor configured to perform steps of a method comprising:

[0008] - receiving a parameter time series (120, a to i) of a measurement of at least one parameter, a test parameter, measured of the test subject after birth of the test subject during a test session;

[0009] - determining a prediction time series (130a to g) of sequential prediction scores for LOS, TLOSPS(ptd), of the test subject, wherein

[0010] - each TLOSPS(ptd) within the prediction time series is determined at a determination time point, ptd;

[0011] - each TLOSPS(ptd) (e.g. 130, a or 130,e) is determined from the at least one parameter time series (e.g. 120, a or 120, a) limited to a parameter time window, PTW, - the PTW spans up to the ptd, and extends back in time according to a size of the PTW;

[0012] - dynamically evaluating the TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd).

[0013] Preferably, the dynamically evaluating comprises:

[0014] - comparing a current TLOSPS(ptd), TLOSPS(ptdc), against a current threshold score, THS(ptdc), wherein the THS(ptdc) is determined from a previous threshold score, THS(ptdp) adjusted by an offset, wherein the offset is determined from:

[0015] - a TsTLOSPS(TTW) that is the prediction time series limited to a threshold time window, TTW, wherein:

[0016] - the TTW has a size in a range of 20 to 40 hours;

[0017] - TTW spans up to but not including the TLOSPS(ptdc) and extends back in time according to the size of the TTW.

[0018] In particular, provided herein is system for dynamic assessment of a prediction of late onset sepsis, LOS, in a test neonatal subject, a test subject, the system comprising a processor configured to perform steps of a method comprising:

[0019] - receiving a parameter time series (120, a to i) of a measurement of at least one parameter, a test parameter, measured of the test subject after birth of the test subject during a test session;

[0020] - determining a prediction time series (130a to g) of sequential prediction scores for LOS, TLOSPS(ptd), of the test subject, wherein

[0021] - each TLOSPS(ptd) within the prediction time series is determined at a determination time point, ptd;

[0022] - each TLOSPS(ptd) (e.g. 130, a or 130,e) is determined from the at least one parameter time series (e.g. 120, a or 120, a) limited to a parameter time window, PTW,

[0023] - the PTW spans up to the ptd, and extends back in time according to a size of the PTW;

[0024] - dynamically evaluating the TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd), which comprises: - comparing a current TLOSPS(ptd), TLOSPS(ptdc), against a current threshold score, THS(ptdc), wherein the THS(ptdc) is determined from a previous threshold score, THS(ptdp) adjusted by an offset, wherein the offset is determined from:

[0025] - a TsTLOSPS(TTW) that is the prediction time series limited to a threshold time window, TTW, wherein:

[0026] - the TTW has a size in a range of 20 to 40 hours; and

[0027] - TTW spans up to but not including the TLOSPS(ptdc) and extends back in time according to the size of the TTW.

[0028] Preferably, the at least one parameter is selected from group comprising Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen (FiO2), temperature, glucose level.

[0029] Preferably, the offset comprises a summation of a negative component and a positive component, wherein:

[0030] - the negative component increases responsive to an increasing proportion of TLOSPS(ptd) values or derivatives thereof in the TsTLOSPS(TTW) that are below a low score value, LSV;

[0031] - the negative component decreases responsive to an increasing proportion of TLOSPS(ptd) values or derivatives thereof in the TsTLOSPS(TTW) that are above the LSV.

[0032] Preferably, the negative component and the positive component are each scaled according to a quantity of discrete TLOSPS’s measured in the TTW.

[0033] Preferably, the parameter time series (120,e to g) is a time series measurement of at least two different parameters.

[0034] Preferably, the parameter time series (120,e to g) is a time series measurement of at least three different parameters that include Heart rate (HR), Saturation (SpO2), and Perfusion index (PFI), and

[0035] - optionally one or more of glucose level and respiratory rate (RR), and

[0036] - optionally one or more of, temperature, admitted oxygen FiO2. Preferably, the method further comprises:

[0037] - triggering an alert when the TLOSPS(ptd) or TLOSPS(ptdc) crosses the threshold score THS(ptd) or TLOSPS(ptdc).

[0038] According to a preferred aspect:

[0039] - a test dataset (606) is generated from:

[0040] - a Ts(ptd)(PTW)TPar (602), wherein a Ts(ptd)(PTW)TPar is the parameter time series for one test paramater, Par, limited to a parameter time window, PTW, or

[0041] - a G-Ts(ptd)(PTW)TPar (602), wherein a G-Ts(ptd)(PTW)TPar is a group of at least two different Ts(ptd)(PTW)TPar’s, each Ts(ptd)(PTW)TPar of the group corresponding to a different test parameter;

[0042] - the PTW spans up to the ptd, and extends back in time according to a size of the PTW;

[0043] - the PTW has a size of 32 to 40 hours;

[0044] - the test dataset (606) is applied to a trained model algorithm (520), wherein the trained model algorithm has been trained using a plurality of training records (504) and is configured to output the TLOSPS(ptd).

[0045] According to a preferred aspect:

[0046] - the trained model algorithm is trained using multiple record datasets (270, a; 270, b; 270, c;),

[0047] - each record dataset is generated from at least one parameter time series (220, a to e) of a measurement of a parameter, a record parameter, measured of a record subject after birth of the record subject during a record session;

[0048] - the at least one parameter time series is limited to a parameter time window (PTW);

[0049] - the PTW spans up to a determination time point ptd, and extends back in time according to a size of the PTW.

[0050] According to a preferred aspect:

[0051] - a record dataset (506) is generated from:

[0052] - a Ts(ptd)(PTW)RPar (602), wherein a Ts(ptd)(PTW)RPar is the parameter time series for one record paramater, Par, limited to a parameter time window, PTW, or - a G-Ts(ptd)(PTW)RPar (602), wherein a G-Ts(ptd)(PTW)RPar is a group of at least two different Ts(ptd)(PTW)RPar’s, each Ts(ptd)(PTW)RPar of the group corresponding to a different record parameter;

[0053] - the PTW spans up to the ptd, and extends back in time according to a size of the PTW;

[0054] - the PTW has a size of 32 to 40 hours;

[0055] - a training record (504) for the training subject is generated from the record dataset (506) and an LOS outcome (508) for the record subject;

[0056] - the multiple training record (504) from multiple different record subject are used to train an untrained model algorithm, thereby generating the trained model algorithm.

[0057] Further provided is a computer program or computer program product having instructions which when executed by a computing device or system cause the computing device or system to perform the method as defined herein.

[0058] Further provided is a computer readable medium having stored thereon instructions which when executed by a computing device or system cause the computing device or system to perform the method as defined herein.

[0059] Figure Legends

[0060] The following description of the figures of specific embodiments of the invention is merely exemplary in nature and is not intended to limit the present teachings, their application or uses.

[0061] FIG. 1 illustrates an exemplary a prediction time series of prediction scores for LOS (TLOSPS(ptd)) at a determination timepoints of 36h, 36.5h, 37h, and 37.5h, wherein each TLOSPS(ptd) is determined from a parameter time series of heart rate (HR) measurements (individual tick marks), where the parameter time series is up to the determination time (ptd), and compared against a dynamically adjusted threshold (THS).

[0062] FIG. 2A illustrates a parameter time series (Ts(ptd=36h)(PTW)TPar) of sequential measurements of one test parameter, HR, for the test subject at multiple different timepoints (pt) within a parameter time window (PTW) of 36h span, at a determination time point (ptd) of 36 hours.

[0063] FIG. 2B illustrates a group of parameter time series (G-Ts(ptd=36h)(PTW)TPar) of sequential measurements of three different test parameters, HR, SpO2 and PFI for the test subject, each parameter time series (Ts(ptd=36h)(PTW)TPar) of the group having the same parameter time window (PTW) span of 36h, and the same determination time point (ptd) of 36 hours.

[0064] FIG. 2C illustrates a parameter time series (Ts(ptd=36h)(PTW)RPar) of sequential measurements of one record parameter, HR, for the record subject at multiple different timepoints (pt) within a parameter time window (PTW) of 36h span, at a determination time point (ptd) of 36 hours.

[0065] FIG. 2D illustrates a group of parameter time series (G-Ts(ptd=36h)(PTW)RPar) of sequential measurements of three different test parameters, HR, SpO2 and PFI for the record subject, each parameter time series (Ts(ptd=36h)(PTW)RPar) of the group having the same parameter time window (PTW) span of 36h, and the same determination time point (ptd) of 36 hours.

[0066] FIGs. 3A to 3C illustrates a prediction time series of predictions of LOSPS, where each LOSPS is determined from a test dataset generated from a G-Ts(ptd)(PTW)TPar. In FIG. 3A, the G-Ts(ptd)(PTW)TPar has parameter time window (PTW) span of 36h, and a determination time point (ptd) of 36 hours, leading to a LOS prediction score (TLOSPS(ptd)) for the test subject at a determination timepoint of 36h. In FIG. 3B, the G-Ts(ptd)(PTW)TPar has parameter time window (PTW) span of 36h, and a determination time point (ptd) of 36.5 hours, leading to a LOS prediction score (TLOSPS(ptd)) for the test subject at a determination timepoint of 36.5h. In FIG. 3C, the G-Ts(ptd)(PTW)TPar has parameter time window (PTW) span of 36h, and a determination time point (ptd) of 37 hours, leading to a LOS prediction score (TLOSPS(ptd)) for the test subject at a determination timepoint of 37h.

[0067] FIGs. 4A to 4C illustrates a record time series of record datasets, where each record datasets is determined from a record dataset generated from a G-Ts(ptd)(PTW)RPar. In FIG. 4A, the G-Ts(ptd)(PTW)RPar has parameter time window (PTW) span of 36h, and a determination time point (ptd) of 36 hours, leading to a record dataset for the record subject at a determination timepoint of 36h. In FIG. 4B, the G-Ts(ptd)(PTW)RPar has parameter time window (PTW) span of 36h, and a determination time point (ptd) of 36.5 hours, leading to a record dataset for the record subject at a determination timepoint of 36.5h. In FIG. 4C, the G- Ts(ptd)(PTW)TPar has parameter time window (PTW) span of 36h, and a determination time point (ptd) of 37 hours, leading to a record dataset for the record subject at a determination timepoint of 37h.

[0068] FIGs. 5A to 5C illustrate determination of the dynamically-adjusted threshold score, THS(ptd). Each round circle is a threshold score (THS) at a determination timepoint (ptd); a diagonal shaded round circle is a threshold score THS(ptdc) at a current determination timepoint; a horizontal shaded round circle is a threshold score THS(ptdp) at a previous determination timepoint. Each square box is a test subject LOS prediction score (TLOSPS) at a determination timepoint (ptd); a diagonal shaded square box is a TLOSPS at a current determination timepoint (TLOSPS (ptdc)); a vertical shaded square box is a TLOSPS at a historic determination timepoint (TLOSPS (ptdh)) which form a prediction time series TsTLOSPS(TTW). In each of FIGs. 5A to 5C, current determination time is advanced by 0.5h. FIG. 6A illustrates an exemplary scheme for training a model algorithm using multiple training records from multiple record subjects.

[0069] FIG. 6B illustrates an exemplary scheme for determining a prediction score for LOS in a test subject using the trained model algorithm.

[0070] FIGs. 7A to 7E shows results of a method of the invention applied to a neonatal subjects x, y, z, alpha and beta.

[0071] Detailed description of invention

[0072] As used herein, the singular forms “a”, “an”, and “the” include both singular and plural referents unless the context clearly dictates otherwise.

[0073] The terms “comprising”, “comprises” and “comprised of” as used herein are synonymous with “including”, “includes”, “containing”, or “contains”, and are inclusive or open-ended and do not exclude additional, non-recited members, elements or method steps. The terms also encompass “constituted of”, “consists in”, “consisting of”, and “consists of”, and also the terms “consisting essentially of”, “consisting essentially in” and “consists essentially of”, which enjoy well-established meanings in patent terminology.

[0074] The recitation of numerical ranges by endpoints includes all intervening values between the lower and upper endpoints, as well as the recited endpoints. Intervening values may be integers or, where applicable, fractions, i.e., more broadly any real numbers such as any rational numbers. This applies to numerical ranges irrespective of whether they are introduced by the expression “from... to...” or the expression “between... and...” or another expression. Any numerical range recited herein is intended to include all sub-ranges subsumed therein. For example, each sub-range between any stated value in a stated range and any other stated value in that stated range is also specifically disclosed. Each sub-range between any stated value in a stated range and either the lower endpoint or the upper endpoint of the stated range is also specifically disclosed. The stated value may be an isolated value or an endpoint of a range subsumed by or overlapping with the stated range. For example, for a stated range with lower endpoint L1 and upper endpoint U1 (i.e., stated range L1-LI1) and a stated sub-range nested within the stated range with lower endpoint L2 and upper endpoint U2 (i.e., stated subrange L2-LI2), also specifically disclosed are the subranges L1-L2, L1-LI2, L2-LI1 , and U2-LI1.

[0075] The terms “about” or “approximately” as used herein when referring to a measurable value such as a parameter, an amount, a temporal duration, and the like, are meant to encompass variations of and from the specified value, such as variations of + / -10% or less, preferably + / - 5% or less, more preferably + / -1 % or less, and still more preferably + / -0.1 % or less of and from the specified value, insofar such variations are appropriate to perform in the disclosed invention. It is to be understood that the value to which the modifier “about” or “approximately” refers is itself also specifically, and preferably, disclosed.

[0076] Furthermore, the terms first, second, third and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order, unless specified. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the invention described herein are capable of operation in other sequences than described or illustrated herein.

[0077] Whereas the terms “one or more” or “at least one”, such as one or more members or at least one member of a group of members, is clear per se, by means of further exemplification, the term encompasses inter alia a reference to any one of said members, or to any two or more of said members, such as, e.g., any >3, >4, >5, >6 or >7 etc. of said members, and up to all said members. In another example, “one or more” or “at least one” may refer to 1 , 2, 3, 4, 5, 6, 7 or more.

[0078] As used herein, the term “and / or” when used in a list of two or more items, means that any one of the listed items can be employed by itself or any combination of two or more of the listed items can be employed. For example, if a list is described as comprising group A, B, and / or C, the list can comprise A alone, B alone, C alone, A and B in combination, A and C in combination, B and C in combination, or A, B, and C in combination.

[0079] The discussion of the background to the invention herein is included to explain the context of the invention. This is not to be taken as an admission that any of the material referred to was published, known, or part of the common general knowledge in any country as of the priority date of any of the claims.

[0080] Throughout this disclosure, various publications, patents and published patent specifications are referenced by an identifying citation. All documents cited in the present specification are hereby incorporated by reference in their entirety. In particular, the teachings or sections of such documents herein specifically referred to are incorporated by reference.

[0081] Unless otherwise defined, all terms used in disclosing the invention, including technical and scientific terms, have the meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. By means of further guidance, term definitions are included to better appreciate the teaching of the invention. When specific terms are defined in connection with a particular aspect of the invention or a particular embodiment of the invention, such connotation or meaning is meant to apply throughout this specification, i.e. , also in the context of other aspects or embodiments of the invention, unless otherwise defined.

[0082] In the following passages, different aspects or embodiments of the invention are defined in more detail. Each aspect or embodiment so defined may be combined with any other aspect(s) or embodiment(s) unless clearly indicated to the contrary. In particular, any feature indicated as being preferred or advantageous may be combined with any other feature or features indicated as being preferred or advantageous.

[0083] Reference throughout this specification to “one embodiment”, “an embodiment” means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout this specification are not necessarily all referring to the same embodiment, but may. Furthermore, the particular features, structures or characteristics may be combined in any suitable manner, as would be apparent to a person skilled in the art from this disclosure, in one or more embodiments. Furthermore, while some embodiments described herein include some but not other features included in other embodiments, combinations of features of different embodiments are meant to be within the scope of the invention, and form different embodiments, as would be understood by those in the art. For example, in the appended claims, any of the claimed embodiments can be used in any combination.

[0084] Similarly, it should be appreciated that in the description of illustrative embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects.

[0085] In the present description of the invention, reference is made to the accompanying drawings that form a part hereof, and in which are shown by way of illustration only of specific embodiments in which the invention may be practiced. Parenthesized or emboldened reference numerals affixed to respective elements merely exemplify the elements by way of example, with which it is not intended to limit the respective elements. Unless otherwise indicated, all figures and drawings in this document are not to scale and are chosen for the purpose of illustrating different embodiments of the invention. In particular the dimensions of the various components are depicted in illustrative terms only, and no relationship between the dimensions of the various components should be inferred from the drawings, unless so indicated.

[0086] While the invention has been described in conjunction with specific embodiments thereof, it is evident that many alternatives, modifications, and variations will be apparent to those skilled in the art in light of the foregoing description. Accordingly, it is intended to embrace all such alternatives, modifications, and variations as follows in the spirit and scope of the appended claims.

[0087] The herein disclosed aspects and embodiments of the invention are further supported by the following non-limiting examples. Provided herein is a method for dynamic assessment of a prediction of late onset sepsis, LOS, in a test neonatal subject, known herein as a test subject. Also provided is a system for carrying out the method.

[0088] The method comprises receiving a parameter time series of measurements of at least one parameter, test parameter, measured of the test subject after birth of the test subject during a test session.

[0089] The at least one parameter is selected from group comprising Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen (FiO2), Temperature, Glucose level.

[0090] The method further comprises determining a prediction time series of prediction scores for LOS, TLOSPS(ptd) of the test subject, wherein each TLOSPS(ptd) within the prediction time series is determined at a determination time (ptd), wherein each TLOSPS(ptd) is determined from the time series of measurements of the at least one parameter up to the determination time (ptd).

[0091] The method further comprises dynamically evaluating the TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd).

[0092] An exemplary scheme is shown in FIG. 1 . FIG. 1 shows a prediction time series (130, a; 130, b; 130, c; 130, d) of prediction scores for LOS (TLOSPS(ptd)) at a determination timepoint of 36h, 36.5h, 37h, and 37.5h. Each TLOSPS(ptd) is determined from the parameter time series (120, a; 120, b; 120, c; 120, d) of measurements of the at least one parameter, in this case, HR, where the parameter time series is up to the determination time (ptd); each parameter measurement is marked with a tick mark. Each TLOSPS(ptd) is compared (cf) with a threshold score THS(ptd) determined for the same determination time (ptd).

[0093] The present methods and systems provide a convenient and rapid method for detection (prediction) of late onset sepsis. The method and system can take advantage of sensors already present in a typical hospital intensive care unit (ICU) environment (e.g. for measurement of Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen (FiO2), Temperature, Glucose level)). The method and system provides dynamic assessment i.e. the LOS prediction score is determined at regularly and up to the current moment, in contrast with the gold standard for detecting LOS in which blood cultures are grown but it takes between 12 and 48 hours to detect microbial growth from the moment of sample extraction. The method and system are particularly useful in a clinical setting for patient triage, i.e. early warning of potential LOS.

[0094] The inventors have found that the dynamically adjusted threshold, i.e. threshold is determined regularly and up to the current moment, can significantly reduce false alarms (FIGs. 7A to E).

[0095] LOS as used herein means late onset sepsis.

[0096] The subject is a neonatal subject i.e. is a newly born infant. The subject is preferably human. The subject may be a test subject, for whom the prediction of LOS and the dynamic assessment there of is to be determined. The subject may be a record subject, for whom the prediction of LOS is known; the record subject is used to train a model algorithm.

[0097] A session is a continuous period during which the (test or training) subject is under observation.

[0098] The session may be a test session where the subject is the test subject, for whom the prediction of LOS and the dynamic assessment thereof is to be determined.

[0099] The session may be a record session where the subject is the record subject, for whom the prediction of LOS is known; the record subject is used to train a model algorithm.

[0100] Typically a (test or record) session starts shortly after birth (e.g. within 1 to 2 hours after birth, or as soon as possible after the moment of birth). It typically ends a few days before discharge in case of good patient outcome; this may range between 3 days to 180, typically 8 days to 30 days after birth. If the dynamic assessment of the prediction of late onset sepsis, LOS, in a test neonatal subject, indicates LOS, the test session may be longer.

[0101] A timepoint (pt) mentioned herein refers to points in time relative to the time of birth of the subject.

[0102] A timepoint (pt) may be a measurement time point - a point in time at which a parameter of the one or more parameters is measured of the subject.

[0103] A timepoint (pt) may be a LOS prediction score time point - a point in time at which the prediction score for LOS (LOSPS) is determined for the test subject A timepoint (pt) may be a threshold timepoint - a point in time at which a threshold score is determined for the subject. The threshold score at a timepoint is known as THS(pt).

[0104] A determination time point (ptd) is a timepoint (pt) at which a determination is or was made of LOSPS or THS.

[0105] A determination time point current (ptdc) is a determination time (ptd) at which a determination is to be made of LOSPS or THS; it is typically called a current timepoint. Where LOSPS or THS are determined at a current timepoint (ptdc), any delay between requesting the LOSPS or THS and determining the LOSPS or THS is limited by computing power. Using a typical processor, a delay is less than 1 minute).

[0106] A determination time point historic (ptdh) is a determination time (ptd) at which was determination was made of LOSPS or THS, and is prior to ptdc; it is typically called a historic timepoint.

[0107] Par as used herein is a parameter (e.g. HR) of the one or more parameters. A parameter (Par) is a bodily measurement of the (test or record) subject, or a measurement of a condition to which the (test or record) subject is subjected to. The parameter (Par) may be heart rate (HR) (BPM), saturation (SpO2), perfusion index (PFI), respiratory rate (RR / min), admitted oxygen FiO2, temperature, or glucose level.

[0108] TPar as used herein is a parameter (e.g. HR) of the one or more parameters measured for the test subject; the TPar is measured in respect of the test subject for whom the LOS prediction score (TLOSPS) and the dynamic assessment there of is to be determined.

[0109] RPar as used herein is a parameter (e.g. HR) of the one or more parameters measured for the record subject; the RPar is measured in respect of the record subject for whom the prediction of LOS is known; the record parameter is used to train a model algorithm.

[0110] Ts(ptd)(PTW)TPar as used herein is a parameter time series of sequential measurements of one test parameter (e.g. HR) of the one or more parameters measured for the test subject at multiple different timepoints (pt) within a parameter time window (PTW). The PTW spans up to the determination timepoint (ptd) (and may include a measurement at the ptd), and extends back in time according to the size of the parameter time window (PTW). A time interval between consecutive timepoints (pt) may be constant (e.g. always 1 min, 5 min, 10 min, 30 mins, 1 hour, 2 hour) or may vary between 2 or more timepoints. The size of the parameter time window (PTW) is preferably 32 to 40 hours, more preferably 35 to 37 hours. The size of the PTW is preferably constant for the whole session. The determination timepoint (ptd) may coincide with a measurement, or may be after a final measurement of the parameter time series. An exemplary Ts(ptd)(PTW)TPar (120, h) is shown in FIG. 2A having a PTW of 36h in length or span and a ptd at 36h for HR parameter; each tick mark represents a discrete parameter measurement. FIG. 1 also shows Ts(ptd)(PTW)Tpar at different ptd’s.

[0111] G-Ts(ptd)(PTW)TPar as used herein is a group of Ts(ptd)(PTW)TPar, comprising multiple different Ts(ptd)(PTW)TPar, each different Ts(ptd)(PTW)TPar corresponding to a different test parameter. The PTW and determination timepoint (ptd) are the same for each and every different parameter time series; the PTW spans up to the determination timepoint (ptd) (and may include a measurement at the ptd), and extends back in time according to the size of the parameter time window (PTW). For instance, a G-Ts(ptd)(PTW)TPar may contain a Ts(ptd)(PTW)TPar for HR and a Ts(ptd)(PTW)TPar for SpO2. The time interval time interval between consecutive measurement timepoints (pt) may be different in each parameter of Ts(ptd)(PTW)TPar. For example, Ts(ptd)(PTW)TPar for Heart rate (HR) may contain measurements made every minute, while Ts(ptd)(PTW)TPar for Glucose may contain measurements made every four hours. An exemplary G-Ts(ptd)(PTW)TPar (120, i) is shown in FIG. 2B; each Ts(ptd)(PTW)TPar of the group has a PTW of 36h in length or span and a ptd at 36h; the group contains three different parameters (HR, SpO2, PFI); each tick mark represents a discrete parameter measurement.

[0112] Ts(ptd)(PTW)RPar as used herein is a parameter time series of sequential measurements of one record parameter (e.g. HR) of the one or more parameters measured for the record subject at multiple different timepoint (pt) within a parameter time window (PTW). The PTW spans up to the determination timepoint (ptd) (and may include a measurement at the ptd), and extends back in time according to the size of the parameter time window (PTW). The time interval between consecutive timepoints (pt) may be constant (e.g. always 1 min, 5 min, 10 min, 30 mins, 1 hour, 2 hour, 4 hours) or may be vary between 2 or more timepoints. The size of the parameter time window (PTW) is preferably 32 to 40 hours, more preferably 35 to 37 hours. The size of the PTW is preferably constant for the whole session. The determination timepoint (ptd) may coincide with a measurement, or may be after a final measurement of the parameter time series. An exemplary Ts(ptd)(PTW)RPar is shown in FIG. 20 having a PTW of 36h in length or span and a ptd at 36h for HR parameter; each tick mark represents a discrete parameter measurement.

[0113] G-Ts(ptd)(PTW)RPar as used herein is a group of Ts(ptd)(PTW)RPar, comprising multiple different Ts(ptd)(PTW)RPar, each different Ts(ptd)(PTW)RPar corresponding to a different record parameter. The PTW and determination timepoint (ptd) are the same for each and every different parameter time series; the PTW spans up to the determination timepoint (ptd) (and may include a measurement at the ptd), and extends back in time according to the size of the parameter time window (PTW). For instance, a G-Ts(ptd)(PTW)RPar may contain a Ts(ptd)(PTW)RPar for HR and a Ts(ptd)(PTW)RPar for SpO2. The time interval time interval between consecutive measurement timepoints (pt) may be different in each parameter of Ts(ptd)(PTW)RPar. For example, Ts(ptd)(PTW)RPar for Heart rate (HR) may contain measurements made every minute, while Ts(ptd)(PTW)RPar for Glucose may contain measurements made every four hours. An exemplary G-Ts(ptd)(PTW)RPar is shown in FIG. 2D; each Ts(ptd)(PTW)RPar of the group has a PTW of 36h in length or span and a ptd at 36h; the group contains three different parameters (HR, SpO2, PFI); each tick mark represents a discrete parameter measurement.

[0114] For a particular model algorithm, the multiple different Ts(ptd)TPar(PTW) TPar parameters in the G-Ts(ptd)TPar(PTW) are found amongst (same as or a subset of) the multiple different Ts(ptd)RPar(PTW) Rpar parameters in the G-Ts(ptd)RPar(PTW). For instance, the multiple different Ts(ptd)RPar(PTW) RPar parameters in the G-Ts(ptd)TPar(PTW) may be HR, SpO2, and PFI in which case the multiple different Ts(ptd)RPar(PTW) TPar parameters in the G- Ts(ptd)RPar(PTW) may be for instance HR and SpO2 (subset of 2 out of 3), or may be HR, SpO2, and PFI (the same).

[0115] For a particular model algorithm, preferably the multiple different Ts(ptd)TPar(PTW) TPar parameters in the G-Ts(ptd)TPar(PTW) are the same as the multiple different Ts(ptd)RPar(PTW) Rpar parameters in the G-Ts(ptd)RPar(PTW). For instance, the multiple different Ts(ptd)TPar(PTW) TPar parameters in the G-Ts(ptd)TPar(PTW) may be HR, SpO2 and PFI, in which case the multiple different Ts(ptd)RPar(PTW) RPar parameters in the G- Ts(ptd)RPar(PTW) are also HR, SpO2, and PFI. THS as used herein is a threshold score.

[0116] THS(ptd) as used herein is a threshold score determined for a determination timepoint.

[0117] THS(ptdc) as used herein is a threshold score determined at a determination timepoint that is a ptdc. Typically, no further determination of a threshold score (THS) has yet been made.

[0118] THS(ptdp) as used herein is a threshold score previously determined at a determination timepoint, wherein it was previously determined at one determination timepoint (dpt) immediately prior to THS(ptdc).

[0119] TLOSPS as used herein is a Test subject late onset sepsis prediction score.

[0120] TLOSPS(ptd) as used herein is a Test subject late onset sepsis prediction score determined at a determination timepoint

[0121] TLOSPS(ptdc) as used herein is a Test subject late onset sepsis prediction score determined at a determination timepoint that is a ptdc. Typically, no further determination of a TLOSPS has yet been made.

[0122] TLOSPS(ptdh) as used herein is a Test subject late onset sepsis prediction score determined at a determination timepoint that is a ptdh.

[0123] TsTLOSPS(TTW) as used herein is a prediction time series of sequential prediction scores for the test subject for late onset sepsis within a threshold time window (TTW). The threshold time window (TTW) is a time window having a size that is preferably 20 to 40 hours, more preferably 25 to 36 hours. The threshold time window (TTW) spans up to but not including the TLOSPS(ptdc) and extends back in time according to the size of the TTW. As such, the TsTLOSPS(TTW) is populated by TLOSPS(ptdh)’s extends back in time according to the size of the TTW. The size of the TTW is preferably constant throughout the session.

[0124] Where different timepoints (pt) are mentioned, it refers to a plurality different timepoints (pt) with a delay between each timepoint. The delay may be the same between the respective timepoints (e.g. always 1 min, 5 min, 10 min, 1 hour, 2 hour), or may vary between two or more timepoints).

[0125] Different time points for determining a LOS prediction score, for measuring a parameter, and for determining a threshold may or may not coincide. A parameter (TPar or Rpar) is a bodily measurement of the (test or record) subject, or a measurement of a condition to which the (test or record) subject is subjected to. The parameter (TPar or Rpar) may be heart rate (HR), saturation (SpO2), perfusion index (PFI), respiratory rate (RR / min), admitted oxygen (FiO2), temperature, or glucose level.

[0126] Heart rate (HR) is a standard bodily measurement of the subject well known in the art. It is typically expressed as beats per minute. It is typically measured using a heart rate sensor, such as a 3-lead ECG sensor but can also be measured with a pulse oximeter. HR may be measured intermittently or continuously.

[0127] Saturation (SpO2), or blood oxygen saturation is a bodily measurement of the subject well known in the art, that is typically expressed in percent (%). It is typically measured using a pulse oximeter. SpO2 may be measured intermittently or continuously.

[0128] Perfusion index (PFI) is a standard bodily measurement of the subject well known in the art. It is a ratio of pulsatile blood flow to non-pulsatile blood flow in the subjects peripheral tissue, such as in a fingertip, toe, or ear lobe. It is typically expressed in percent (%). It is typically measured using a pulse oximeter. PFI may be measured intermittently or continuously.

[0129] Respiratory rate (RR), is a standard bodily measurement of the subject well known in the art. It expressed as a quantity of breaths per minute. It is typically measured using a 3-lead ECG sensor. RR may be measured intermittently or continuously.

[0130] Admitted oxygen (FiO2) is a standard measurement well known in the art. It refers to a concentration of oxygen in the air or in a gas mixture fed to the subject under assisted ventilation. It is typically expressed in percent (%). For passive breathing, atmospheric air admitted oxygen is typically 21 %. For ventilator-assisted breath, the concentration of oxygen in the gas mix fed to the patient can be read out. FiO2 may be measured intermittently or continuously.

[0131] Temperature is a standard measurement well known in the art. It refers to a core temperature of the subject. It is typically expressed in centigrade, Kelvin or Fahrenheit. It is typically measured using a rectal-positioned thermal probe, an armpit-positioned thermal probe, or an infrared tympanic device. Temperature may be measured intermittently or continuously.

[0132] Glucose level is a standard measurement well known in the art. It refers to a concentration of glucose in the blood of the subject. It is typically expressed in mmol / L. It may be measured intermittently by taking regular blood samples. The at least one parameter (TPar or RPar) is selected from group comprising Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen FiO2, Temperature, Glucose level to determine the LOS prediction score (TLOSPS) and / or used to train the model algorithm. Preferably, at least two parameters are selected from the aforementioned group. Preferably, all the parameters are selected from the aforementioned group.

[0133] The at least one parameter (TPar or RPar) used to determine the LOS prediction score (TLOSPS) or used to train the model algorithm preferably comprises at least HR, SpO2 and PFI.

[0134] The at least one parameter (TPar or RPar) used to determine the LOS prediction score (TLOSPS) or used to train the model algorithm preferably comprises at least HR, SpO2 and PFI, and optionally one or more of Glucose level, RR and optionally one or more of temperature and FiO2.

[0135] The at least one parameter used to determine the LOS prediction score (TLOSPS) or used to train the model algorithm preferably comprises at least HR, SpO2, PFI, and

[0136] - one or more of glucose level, RR, and

[0137] - optionally one or more of Temperature, FiO2.

[0138] The at least one parameter used to determine the LOS prediction score (TLOSPS) or used to train the model algorithm may comprise HR, SpO2, PFI, RR, FiO2, temperature, and glucose level.

[0139] According to one aspect, for a G-Ts(ptd)(PTW)TPar comprising multiple different Ts(ptd)(PTW)TPar’s, each different Ts(ptd)(PTW)TPar corresponding to a different test parameter, the different test parameters are selected from Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen FiO2, Temperature, Glucose level to determine the LOS prediction score (TLOSPS(ptdc) and / or used to train the model algorithm. Preferably, at least two test parameters are selected from the aforementioned group. Preferably, all the test parameters are selected from the aforementioned group.

[0140] The different test parameters preferably include at least HR, SpO2 and PFI.

[0141] The different test parameters preferably include at least HR, SpO2 and PFI, and optionally one or more of RR and FiO2 and optionally one or more of Temperature, Glucose level. The different test parameters preferably include HR, SpO2, PFI, and

[0142] - one or more of RR, FiO2, and

[0143] - optionally one or more of Temperature, Glucose level.

[0144] According to one aspect, for a G-Ts(ptd)(PTW)RPar comprising multiple different Ts(ptd)(PTW)RPar’s, each different Ts(ptd)(PTW)RPar corresponding to a different record parameter, the different record parameters are selected from Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen FiO2, Temperature, Glucose level to determine the LOS prediction score (TLOSPS(ptdc) and / or used to train the model algorithm. Preferably, at least two record parameters are selected from the aforementioned group. Preferably, all the record parameters are selected from the aforementioned group.

[0145] The different record parameters preferably include at least HR, SpO2 and PFI.

[0146] The different record parameters preferably include at least HR, SpO2 and PFI, and optionally one or more of RR and FiO2.

[0147] The different record parameters preferably include HR, SpO2, PFI, and

[0148] - one or more of RR, FiO2, and

[0149] - optionally one or more of Temperature, Glucose level.

[0150] The method comprises dynamically evaluating a TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd). More in particular, a current TLOSPS(ptd), TLOSPS(ptdc), is compared against a current threshold score, THS(ptdc), and the THS(ptdc) is determined from a previous threshold score, THS(ptdp).

[0151] The THS(ptdc) is equal to the THS(ptdp) adjusted by an offset. The offset is determined from the TsTLOSPS(TTW).

[0152] This is illustrated schematically in FIGs. 5 A to C. In FIG. 5A, four TLOSPS(ptd) have been determined at a determination timepoint (square boxes), and the timepoint of the TLOSPS(ptdc) (diagonal shaded square box) is 37.5h. To evaluate the TLOSPS(ptdc), a THS (ptdc) (diagonal shaded round circle) is calculated. The THS(ptdc) is calculated from the THS(ptdp) (horizontal shaded round circle) adjusted by an offset. The offset is calculated from the prediction time series TsTLOSPS(TTW) i.e. values of TLOSPS(ptdh) within a 24 hours TTW window. In FIG. 5B, time has advanced by 0.5h. Five TLOSPS(ptd) have been determined (square boxes) and the timepoint of the TLOSPS(ptdc) (diagonal shaded square box) is 38h. To evaluate the TLOSPS(ptdc), a new THS (ptdc) (diagonal shaded round circle) is calculated. The THS(ptdc) is calculated from the THS(previous) (horizontal shaded round circle) adjusted by an offset. The offset is calculated from the prediction time series TsTLOSPS(TTW) i.e. values of TLOSPS(ptdh) within a 24 hours TTW window.

[0153] In FIG. 5C, time has advanced again by 0.5h. Six TLOSPS(ptd) have been determined (square boxes) and the timepoint of the TLOSPS(ptdc) (diagonal shaded square box) is 38.5h. To evaluate the TLOSPS(ptdc), a new THS (ptdc) (diagonal shaded round circle) is calculated. The THS(ptdc) is calculated from the THS(ptdp) (horizontal shaded round circle) adjusted by an offset. The offset is calculated from the prediction time series TsTLOSPS(TTW) i.e. values of TLOSPS(ptdh) within a 24 hours TTW window.

[0154] The offset comprises a combination (by summation) of a negative component and a positive component.

[0155] The negative component increases i.e. the negative component becomes more negative, the THS(ptdc) lowers) responsive to an increasing proportion of TLOSPS(ptdh) values or derived values therefrom in the TsTLOSPS(TTW), that are below a low score value, LSV. An example of the derived values is provided below herein.

[0156] The positive component increases (positive component increases becomes more positive, the THS(ptdc) rises) responsive to an increasing proportion of TLOSPS(ptdh) values or derived values therefrom in the TsTLOSPS(TTW) that are above the low score value, LSV. An example of the derived values is provided below herein.

[0157] Optionally the negative component and the positive component are both scaled according to a quantity of discrete TLOSPS(ptdh) measured in the TTW. The scaling is greater for the positive component {e.g. 2x, 3x, 4x, 5x greater) compared with the negative component. The low score value, LSV, is a threshold set by the user that is equal to or less than 20% of the scale (e.g. 20% or 0.2; 15% or 0.15; 10% or 0.1).

[0158] According to a particular aspect:

[0159] THS(ptdc) = THS(ptdp) - (Propjow * NegCompScaling) + ((1-Prop_low) * increase_factor * PosCompScaling)

[0160] Where:

[0161] Propjow = proportion of TLOSPS(ptdh) values of the TsTLOSPS(TTW) below the LSV, wherein the proportion is calculated by:

[0162] - applying a rolling threshold time window (RTTW) to the multiple TLOSPS(ptdh) within the TsTLOSPS(TTW), wherein the rolling threshold time window (RTTW) is smaller in size than the threshold time window (TTW), preferably is 0.2 to 0.7 of the threshold time window (TTW). Preferably the rolling threshold time window (RTTW) is 8 to 14 hours in length, compared with threshold time window (TTW) being preferably 20 to 28 hours in length. The group of TLOSPS(ptdh) in the rolling threshold time window (RTTW) at an index position are known as G-TLOSPS(ptdh)(RTTWi),

[0163] - determining from each G-TLOSPS(ptdh)(RTTWi), a maximum value of TLOSPS(ptdh), thereby obtaining multiple maximum values of TLOSPS(ptdh) (called MMV herein), one for each G-TLOSPS(ptdh)(RTTWi),

[0164] - determining, a proportion of TLOSPS(ptdh) values within the MMV that are below the LSV,

[0165] - optionally, the values of TLOSPS(ptdh) each G-TLOSPS(ptdh)(RTTWi) may be weighted such that values of TLOSPS(ptdh) earlier in the G-TLOSPS(ptdh)(RTTWi) have a higher weighting than that values of TLOSPS(ptdh) later in the G- TLOSPS(ptdh)(RTTWi). The difference in weighting between TLOSPS(ptdh) at the start of the RTTWi and the end of the RTTWi may be linear. In other words, the values of TLOSPS(ptdh) within each G-TLOSPS(ptdh)(RTTWi) may be weighted with time, and the weighting over time is linear and causes a decrease.

[0166] NegCompScaling = 0.45 / ( 5 * quantity of TLOSPS(ptdh) values in the TTW). The value of NegCompScaling is smaller compared with PosCompScaling for the same quantity TLOSPS(ptdh) values within the TTW. lncrease_factor is a multiplication factor for high probabilities resulting in faster threshold increase. Increase factor is TLOSPS(ptdp) * 10, capped at 1 (where TLOSPS(ptdp) * 10 < 1) and capped at 10 (where TLOSPS(ptdp) * 10 > 10).

[0167] PosCompScaling = 0.45 / (quantity of historic TLOSPS determined within the time window (TW). The value of PosCompScaling is larger compared with NegCompScaling for the same quantity TLOSPS(ptdh) values within the TTW;

[0168] The method determines a LOS prediction score. The LOS prediction score may be a value on a scale having a first limit (e.g. 0) and a second limit (e.g. 1 , 10, 100). One limit (e.g. first limit) is lower than the other limit (e.g. second limit). One limit (e.g. first limit) is indicative of a low likelihood of LOS, and the other limit (e.g. second) is indicative of a high likelihood of LOS.

[0169] The prediction score (TLOSPS(ptd)) at determination time point (ptd) is determined from the parameter time series of measurements of the at least one test parameter up to the determination time point (ptd) during the test session.

[0170] In particular, the parameter time series of at least one test parameter is used to generate a test dataset, the test dataset is applied a trained model algorithm. An output of the trained model algorithm is the LOS prediction score. This is described in more detail elsewhere herein.

[0171] A dataset may be a test dataset generated from the Ts(ptd)(PTW)TPar i.e. from multiple parameter time series of measurements of a single parameter measured from the single test subject for whom the prediction of LOS and the dynamic assessment thereof is to be determined. Alternatively a dataset may be a test dataset generated from the G- Ts(ptd)(PTW)TPar i.e. from the multiple parameter time series of measurements of at least two parameters measured from the single test subject for whom the prediction of LOS and the dynamic assessment thereof is to be determined. A test dataset is generated for a ptd.

[0172] A prediction time series of predictions of LOSPS, where each LOSPS is determined from a test dataset generated from a G-Ts(ptd)(PTW)TPar is illustrated schematically in FIG. 3A to C. In FIGs. 3 A to C, the parameter time window (PTW) is the same (36 hours), and three parameters are measured (HR, SpO2 and PFI) for a test subject. In FIG. 3A, a parameter time series of measurements is made for each of HR, SpO2 and PFI for a test subject for a 36h parameter time window ending and including a measurement up to 36 hours. Each parameter time series is labelled Ts(ptd=36h)(PTW)TPar (HR), Ts(ptd=36h)(PTW)TPar (SpO2), or Ts(ptd=36h)(PTW)TPar (PFI); each tick mark represent each measurement. The respective parameter time series form a group of parameter time series G-Ts(ptd=36h)(PTW)TPar (120, e). The G-Ts(ptd =36h)(PTW)TPar (120, e) is used to generate a test dataset (170, e) from which a test LOS prediction score (TLOSPS(t=36h)) (130,e) for the 36 hours time point is determined for the test subject. FIG. 3B is similar to FIG. 3A, except time has advanced by 0.5 hours, and the group (120, f) of parameter time series of measurements contains measurements of HR, SpO2 and PFI for the test subject for the 36h parameter time window ending and including a measurement at 36.5 hours. The test LOS prediction score (TLOSPS(t=36.5h)) (130, f) for the 36.5 hours time point is determined from the test dataset (170, f) for the test subject. FIG. 3C is similar to FIG. 3B, except time has advanced by a further 0.5 hours, and the group (120, g) of parameter time series of measurements contains measurements of HR, SpO2 and PFI for the test subject for the 36h parameter time window ending and including a measurement at 37 hours. The test LOS prediction score (TLOSPS(t=37h)) (130, g) for the 37 hours time point is determined from the test dataset (170, g) for the test subject.

[0173] A dataset may be a record dataset generated from the Ts(ptd)(PTW)RPar i.e. from the multiple parameter time series of measurements of one parameter measured from the single record subject for whom the prediction of LOS is known; the record dataset is used to train a model algorithm.

[0174] Alternatively, a dataset may be a record dataset generated from the G-Ts(ptd)(PTW)RPar / .e. from the parameter time series of measurements of at least two parameters measured from the single record subject for whom the prediction of LOS is known; the record dataset is used to train a model algorithm.

[0175] A record dataset is generated for a ptd. A record time series contains sequential record datasets for a single record subject measured at a different ptd’s (e.g. at 36h, 36.5h, 37h etc). The record time series in particular is used to train a model algorithm. A record time series of record datasets, each record dataset being generated from a G- Ts(ptd)(PTW)RPar is illustrated schematically in FIG. 4 A to C. FIGs. 4 A to C, the parameter time window (PTW) is the same (36 hours), and three parameters are measured (HR, SpO2 and PFI) for a record subject. In FIG. 4A, a parameter time series of measurements is made for each of HR, SpO2 and PFI for a record subject for a 36h parameter time window ending and including a measurement up to 36 hours. Each parameter time series is labelled Ts(ptd=36h)(PTW) RPar (HR), Ts(ptd=36h)(PTW) RPar (SpO2), or Ts(ptd=36h)(PTW) RPar (PFI); each tick mark represents each measurement. The respective parameter time series form a group of parameter time series G-Ts(ptd=36h)(PTW) RPar (220, a). The G-Ts(ptd =36h)(PTW) RPar (220, a) is used to generate a record dataset (270, a) for the 36 hours time point is determined for the record subject. FIG. 4B is similar to FIG. 4A, except time has advanced by 0.5 hours, and the group (220, b) of parameter time series of measurements contains measurements of HR, SpO2 and PFI for record subject for the 36h parameter time window ending and including a measurement at 36.5 hours. The record dataset (270, b) for the 36.5 hours time point is determined for the record subject. FIG. 4C is similar to FIG. 4B, except time has advanced by a further 0.5 hours, and the group (220, c) of parameter time series of measurements contains measurements of HR, SpO2 and PFI for the record subject for the 36h parameter time window ending and including a measurement at 37 hours. The record dataset (270, c) for the 37 hours time point is determined for the record subject. The record time series (280) contains multiple record datasets, each record dataset being generated from a G-Ts(ptd)(PTW)RPar. The multiple record datasets are determined at different ptd’s.

[0176] Generation a (test or record) dataset comprises extracting one or more metrics from each Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar.

[0177] Where a parameter Par is heart rate (HR), the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0178] - mean / median HR over the parameter time series within the PTW;

[0179] - max / min HR over the parameter time series within the PTW;

[0180] - standard deviation of the HR over the parameter time series within the PTW; - counts of HR over time across the parameter time series within the PTW; more in particular a quantity of HR values above or below a medically relevant threshold (e.g. above 180) across the parameter time series within the PTW;

[0181] - approximate entropy of the HR over the parameter time series within the PTW.

[0182] Where a parameter Par is Saturation (SpO2), the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0183] - mean / median SpO2 across the parameter time series within the PTW;

[0184] - max / min SpO2 across the parameter time series within the PTW;

[0185] - standard deviation of the SpO2 across the parameter time series within the PTW;

[0186] - counts of SpO2 across the parameter time series within the PTW; more in particular a quantity of SpO2 values above or below a medically relevant threshold (e.g. below 80) across the parameter time series within the PTW;

[0187] - approximate entropy of the SpO2 across the parameter time series within the PTW.

[0188] Where a parameter Par is Perfusion index (PFI), the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0189] - mean / median PFI across the parameter time series within the PTW;

[0190] - max / min PFI across the parameter time series within the PTW;

[0191] - standard deviation of the PFI across the parameter time series within the PTW;

[0192] - counts of PFI across the parameter time series within the PTW; more in particular a quantity of PFI values above or below a medically relevant threshold across the parameter time series within the PTW;

[0193] - approximate entropy of the PFI across the parameter time series within the PTW.

[0194] Where a parameter Par is Respiratory rate (RR), the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0195] - mean / median RR across the parameter time series within the PTW;

[0196] - max / min RR across the parameter time series within the PTW;

[0197] - standard deviation of the RR across the parameter time series within the PTW;

[0198] - counts of RR across the parameter time series within the PTW; more in particular a quantity of RR values above or below a medically relevant threshold across the parameter time series within the PTW; - approximate entropy of the RR across the parameter time series within the PTW.

[0199] Where a parameter is admitted oxygen (FiO2), the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0200] - mean / median FiO2 across the parameter time series within the PTW;

[0201] - max / min FiO2 across the parameter time series within the PTW;

[0202] - standard deviation of the FiO2 across the parameter time series within the PTW;

[0203] - counts of FiO2 across the parameter time series within the PTW; more in particular a quantity of FiO2 values above or below a medically relevant threshold across the parameter time series within the PTW;

[0204] - approximate entropy of the FiO2 across the parameter time series within the PTW.

[0205] Where a parameter is temperature, the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0206] - mean / median temperature across the parameter time series within the PTW;

[0207] - max / min temperature across the parameter time series within the PTW;

[0208] - standard deviation of the temperature across the parameter time series within the PTW;

[0209] - counts of temperature across the parameter time series within the PTW; more in particular a quantity of temperature values above or below a medically relevant threshold across the parameter time series within the PTW;

[0210] - approximate entropy of the temperature across the parameter time series within the PTW.

[0211] Where a parameter is glucose level, the metrics extracted from the corresponding Ts(ptd)(PTW)TPar or Ts(ptd)(PTW)RPar preferably include one or more (e.g. all) of:

[0212] - mean / median temperature across the parameter time series within the PTW;

[0213] - max / min temperature across the parameter time series within the PTW;

[0214] - standard deviation of the temperature across the parameter time series within the PTW; more in particular a quantity of glucose level values above or below a medically relevant threshold across the parameter time series within the PTW;

[0215] - counts of temperature across the parameter time series within the PTW; - approximate entropy of the temperature across the parameter time series within the PTW.

[0216] An untrained model algorithm is trained using multiple training records. A (single) training record comprises:

[0217] - a record series of time-sequential record datasets for the record subject; and

[0218] - an outcome of LOS, a LOS outcome, of the same record subject for the same record session.

[0219] The LOS outcome of the record subject is binary (e.g. yes, no; 1 , 0; 0, 100) according to a presence or an absence of LOS in the record subject.

[0220] The training comprises adjusting (one or more variables of) the untrained model algorithm such that the record dataset approaches the LOS outcome of the training record. The result of the training is a trained model algorithm.

[0221] An example of a scheme (500) for training the model algorithm is shown in FIG. 6A. Multiple training records (510) are applied to a untrained model algorithm (512). Using methods known in the art, one or more variables of the untrained model algorithm is adjusted such that the record dataset approaches the LOS outcome, thereby generating a trained model algorithm (520). Each training record (504) of the multiple training records (510) is for one record subject for whom the LOS outcome (508) is known, and contains a record time series (506) generated from the multiple Ts(ptd)(PTW)RPar or G(ptd)-Ts(PTW)RPar each at a different ptd.

[0222] The trained model algorithm (520) is used to generate a LOS prediction score for the test subject from one parameter time series of one parameter (T s(ptd)(PTW)TPar) or from a group of parameter time series of different parameters (G-Ts(ptd)(PTW)TPar). An example of a scheme (600) for generate a LOS prediction score (620) for the test subject at a current determination time point (ptdc) from the trained model algorithm (520) is shown in FIG. 6B. A Ts(ptd)(PTW)TPar or a G-Ts(ptd)(PTW)TPar of the test subject at the ptdc (602) is used to generated a test dataset (606). The test dataset (606) is applied to the trained model algorithm (520), and the trained model algorithm (520) outputs a test LOS prediction score for the ptdc for the test subject (TLOSPS(ptdc)) (620). The TLOSPS(ptdc) (620) is compared with the threshold score at the ptdc, thereby evaluating (630) the TLOSPS(ptdc).

[0223] The model algorithm may be a linear correlation (optimised), a machine learning model (e.g. deep learning, CNN) (trained), or any type of model that is refined, optimized, trained or solved using the multiple training records. The model algorithm may be an age-aware model algorithm.

[0224] In general, the training of model algorithms and use of the trained model algorithm is understood by the person skilled in the art. See, for instance, An introduction to statistical learning by James, Gareth, Daniela Witten, Trevor Hastie, and Robert Tibshirani, 2013; or, Machine learning in healthcare informatics, ISBN : 978-3-642-40016-2, 2014.

[0225] The method may generate an output. The output may numeric, graphical, or an alert as described elsewhere herein. An example of an output is an indication of THS(ptd) or THS(ptdc) and TLOSPS(ptd) or TLOSPS(ptdc) at the same ptd or ptdc. The indication may be expressed numerically (e.g. side-by-side comparison, or a scalar value representative of a difference), graphically (e.g. chart plot), using a visual code (e.g. a colour scale (e.g. green to red) or icon scale (representing low, medium or high risk) representative of a closeness to the TLOSPS(ptd) or TLOSPS(ptdc) to the THS(ptd) or THS(ptdc).

[0226] As mentioned, the method further comprises dynamically evaluating the TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd).

[0227] The method may further comprising triggering an alert when the TLOSPS(ptd) or TLOSPS(ptdc) crosses the threshold score (e.g. rises above), THS(ptd) or THS(ptdc). The alert may be any, for instance, an audible and / or visual alert. Examples of an audible alert include a sound generated by a loudspeaker, buzzer, bell or any other sound-generating device. Examples of a visual alert include changing a property of a light (e.g. from off to on, a change in colour, a change in intensity, a change in flashing frequency), displaying an indication on a computer screen. The alert may include sending a message to a communication device such as a telephone, a mobile device (mobile phone, smart phone, tablet etc). The method or system may produce an output.

[0228] The method or system may produce an output that is:

[0229] - displayed on a screen, or

[0230] - saved to a file, or

[0231] - printed.

[0232] Based on the output may method or system may suggest one or more

[0233] - actions steps, or

[0234] - recommendations, or

[0235] - instructions.

[0236] Based on the output may method or system may trigger an alert.

[0237] The output may be transmitted over a communication channel (to a further computing device) The output may be transmitted to one or more users over a computer network in real time, so that each user has access to real time information of the dynamic assessment or output.

[0238] The present method is a computer-implemented method.

[0239] Further provided is a computing device or system configured for performing the method as described herein.

[0240] The system comprises circuitry configured to perform the method of the invention. Typically the circuitry comprises a processor and a memory. Typically the circuitry comprises a processor and a memory. It may include one or more data input ports for receiving data and / or one or more data input ports for sending data (output).

[0241] Further provided is a computer program or computer program product having instructions which when executed by a computing device or system cause the computing device or system to perform the method as described herein. Further provided is a computer readable medium having stored thereon instructions which when executed by a computing device or system cause the computing device or system to perform the method as described herein.

[0242] The processor may be a part of, or method may be performed using a standard computer system such as an Intel Architecture IA-32 based computer system 2, and implemented as programming instructions of one or more software modules stored on non-volatile (e.g., hard disk or solid-state drive) storage associated with the corresponding computer system. However, it will be apparent that at least some of the steps of any of the described processes could alternatively be implemented, either in part or in its entirety, as one or more dedicated hardware components, such as gate configuration data for one or more field programmable gate arrays (FPGAs), or as application-specific integrated circuits (ASICs), for example. The processor may be a part of, or method may be performed using a single standard computer system and / or a part of system of networked computers.

[0243] The method, computing device, or system produces an output. The output may be a value, a signal representative of the value. The output may be displayable or displayed on a display (e.g. computer screen, mobile device (smart phone tablet)). The output may be used to control (directly or indirectly) another device of system.

[0244] Example

[0245] Example 1

[0246] An XGBboost machine learning model algorithm (Chen, Tianqi, and Carlos Guestrin. "Xgboost: A scalable tree boosting system." Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, 2016; Ogunleye, Adeola, and Qing-Guo Wang. "XGBoost model for chronic kidney disease diagnosis." IEEE / ACM transactions on computational biology and bioinformatics 17.6 (2019): 2131-2140) was trained in accordance with the present disclosure using 7841 training records from 1960 neonatal record subjects according to a method, wherein the test parameters were all of heart rate (HR), saturation (SpO2), perfusion index (PFI), respiratory rate (RR), admitted oxygen FiO2, temperature, and glucose level, and the outcome of LOS was known from diagnostic (seroculture) testing. A trained model algorithm was obtained. According to an analysis of the trained model algorithm, a relative importance of each parameter could be determined, as set out in Table 1 below.

[0247] Table 1 : Relative importance of each parameter in the trained model algorithm according to the present disclosure.

[0248] Example 2

[0249] Parameter time series of five neonatal subjects x, y, z, alpha and beta were applied to the trained model algorithm of Example 1 in accordance with the present disclosure from which a prediction time series was obtained over a span of 7 to 8 days. Each TLOSPS(pdt) was compared against a THS(pdt) according to the present disclosure.

[0250] FIG. 7A shows the result in respect of neonatal subject x shown by diagnostic test to be suffering from LOS. The TLOSPS(pdt) showed a peak at ~8 days that crossed the THS(pdt). A diagnostic (seroculture) test for LOS was also performed at ~8 days and was positive for LOS.

[0251] FIG. 7B shows the result in respect of neonatal subject y shown by diagnostic (seroculture) test to be suffering from LOS. The TLOSPS(pdt) showed a peak at ~7 days that crossed the THS(pdt). A diagnostic test for LOS was also performed at ~7 days and was positive for LOS.

[0252] FIG. 7C shows the result in respect of neonatal subject z shown by an initial diagnostic test not to be suffering from LOS and shown by a later diagnostic (seroculture) test to be suffering from LOS. The TLOSPS(pdt) showed a peak at ~3.4 days which did not cross the TLOSPS(pdt); diagnostic (seroculture) test for LOS was also performed at ~3.4 days and was negative for LOS. Later, the TLOSPS(pdt) showed a peak at ~8 days which did cross the TLOSPS(pdt); diagnostic test for LOS was also performed at ~8 days and was positive for LOS.

[0253] FIG. 7D shows the result in respect of neonatal subject alpha shown by diagnostic (seroculture) test to be suffering from LOS. The TLOSPS(pdt) showed a peak at ~7 days that crossed the THS(pdt). A diagnostic test for LOS was also performed at ~7 days and was positive for LOS.

[0254] FIG. 7E shows the result in respect of neonatal subject beta showing 2 peaks neither of which cross the TLOSPS(pdt), and showing by diagnostic (seroculture) test no LOS on both occasions. The TLOSPS(pdt) showed a peak at ~5.5 days which did not cross the TLOSPS(pdt); diagnostic (seroculture) test for LOS was performed at ~5.8 days and was negative for LOS. Later, the TLOSPS(pdt) showed a peak at ~7.8 days which also did not cross the TLOSPS(pdt); diagnostic test for LOS was performed at ~7.8 days and was negative for LOS.

Claims

Claims1. A system for dynamic assessment of a prediction of late onset sepsis, LOS, in a test neonatal subject, a test subject, the system comprising a processor configured to perform steps of a method comprising:- receiving a parameter time series (120, a to i) of a measurement of at least one parameter, a test parameter, measured of the test subject after birth of the test subject during a test session;- determining a prediction time series (130a to g) of sequential prediction scores for LOS, TLOSPS(ptd), of the test subject, wherein- each TLOSPS(ptd) within the prediction time series is determined at a determination time point, ptd;- each TLOSPS(ptd) (e.g. 130, a or 130,e) is determined from the at least one parameter time series (e.g. 120, a or 120, a) limited to a parameter time window, PTW,- the PTW spans up to the ptd, and extends back in time according to a size of the PTW;- dynamically evaluating the TLOSPS(ptd), by comparing the TLOSPS(ptd) against a dynamically-adjusted threshold score, THS(ptd), which comprises:- comparing a current TLOSPS(ptd), TLOSPS(ptdc), against a current threshold score, THS(ptdc), wherein the THS(ptdc) is determined from a previous threshold score, THS(ptdp) adjusted by an offset, wherein the offset is determined from:- a TsTLOSPS(TTW) that is the prediction time series limited to a threshold time window, TTW, wherein:- the TTW has a size in a range of 20 to 40 hours; and- TTW spans up to but not including the TLOSPS(ptdc) and extends back in time according to the size of the TTW.

2. The system according to claim 1 , wherein the at least one parameter is selected from group comprising Heart rate (HR), Respiratory rate (RR), Perfusion index (PFI), Saturation (SpO2), Admitted oxygen (FiO2), temperature, glucose level.

3. The system according to claim 1 or 2, wherein the offset comprises a summation of a negative component and a positive component, wherein:- the negative component increases responsive to an increasing proportion of TLOSPS(ptd) values or derivatives thereof in the TsTLOSPS(TTW) that are below a low score value, LSV;- the negative component decreases responsive to an increasing proportion of TLOSPS(ptd) values or derivatives thereof in the TsTLOSPS(TTW) that are above the LSV.

4. The system according to claim 3, wherein the negative component and the positive component are each scaled according to a quantity of discrete TLOSPS’s measured in the TTW.

5. The system according to any one of claims 1 to 4, wherein the parameter time series (120, e to g) is a time series measurement of at least two different parameters.

6. The system according to any one of claims 1 to 5, wherein the parameter time series (120, e to g) is a time series measurement of at least three different parameters that include Heart rate (HR), Saturation (SpO2), and Perfusion index (PFI), and- optionally one or more of glucose level and respiratory rate (RR), and- optionally one or more of, temperature, admitted oxygen FiO2.

7. The system according to any one of claims 1 to 6, wherein the method further comprises:- triggering an alert when the TLOSPS(ptd) or TLOSPS(ptdc) crosses the threshold score THS(ptd) or TLOSPS(ptdc).

8. The system according to any one of claims 1 to 7, wherein:- a test dataset (606) is generated from:- a Ts(ptd)(PTW)TPar (602), wherein a Ts(ptd)(PTW)TPar is the parameter time series for one test paramater, Par, limited to a parameter time window, PTW, or- a G-Ts(ptd)(PTW)TPar (602), wherein a G-Ts(ptd)(PTW)TPar is a group of at least two different Ts(ptd)(PTW)TPar’s, each Ts(ptd)(PTW)TPar of the group corresponding to a different test parameter;- the PTW spans up to the ptd, and extends back in time according to a size of the PTW;- the PTW has a size of 32 to 40 hours;- the test dataset (606) is applied to a trained model algorithm (520), wherein the trained model algorithm has been trained using a plurality of training records (504) and is configured to output the TLOSPS(ptd).

9. The system according to claim 8, wherein:- the trained model algorithm is trained using multiple record datasets (270, a; 270, b; 270, c;),- each record dataset is generated from at least one parameter time series (220, a to e) of a measurement of a parameter, a record parameter, measured of a record subject after birth of the record subject during a record session;- the at least one parameter time series is limited to a parameter time window (PTW);- the PTW spans up to a determination time point ptd, and extends back in time according to a size of the PTW.

10. The system according to claim 9, wherein:- a record dataset (506) is generated from:- a Ts(ptd)(PTW)RPar (602), wherein a Ts(ptd)(PTW)RPar is the parameter time series for one record paramater, Par, limited to a parameter time window, PTW, or- a G-Ts(ptd)(PTW)RPar (602), wherein a G-Ts(ptd)(PTW)RPar is a group of at least two different Ts(ptd)(PTW)RPar’s, each Ts(ptd)(PTW)RPar of the group corresponding to a different record parameter;- the PTW spans up to the ptd, and extends back in time according to a size of the PTW;- the PTW has a size of 32 to 40 hours;- a training record (504) for the training subject is generated from the record dataset (506) and an LOS outcome (508) for the record subject;- the multiple training record (504) from multiple different record subject are used to train an untrained model algorithm, thereby generating the trained model algorithm.

11. A computer program or computer program product having instructions which when executed by a computing device or system cause the computing device or system to perform the method as defined in any one of claims 1 to 10.

12. A computer readable medium having stored thereon instructions which when executed by a computing device or system cause the computing device or system to perform the method as defined in any one of claims 1 to 10.

Images