Selected laboratory projects.

Whole-blood automation on an existing liquid handler

• Situation

  Clinical pharmacology at Karolinska University Hospital, a department of three accredited laboratories: therapeutic drug monitoring, drugs of abuse, and the Swedish doping control laboratory. The therapeutic drug monitoring laboratory owned a Hamilton STARlet and had five people preparing whole blood by hand for immunosuppressant and phosphatidylethanol analysis, because whole blood separates while it stands and the deck was considered unable to handle it.

• My role

  Biomedical scientist on staff, from 2015 the department’s main programmer of its Hamilton liquid handlers. I developed the method, put it on the deck and kept it in routine.

• What changed

  I measured how long a sample stays mixed after agitation and showed that pipetting below the plasma layer reads the same as the mixed sample. The method was built around that window, with printed tube adapters so samples went on in the tubes they arrived in.

• Outcome

  One operator prepared what five people had prepared by hand, five plates a day, and the laboratory’s own turnaround figures came back at about half. No instrument was bought. Results for that workflow, in that laboratory.

• Evidence

  Poster P25-6-7 at IATDMCT 2017 in Kyoto, first author, which carries the figures below; the department’s subsequent purchase of five more decks.

• Related service

  Laboratory automation and laboratory workflow.

The method, and the measurement behind it

• 5 → 1One operator preparing work that five had prepared by hand
• ~½Turnaround, on the laboratory’s own figures
• 0Instruments purchased; the Hamilton deck was already there

Whole blood was where the volume and the value were, immunosuppressants for transplant patients and phosphatidylethanol as an alcohol marker, and it was also the sample the deck was not supposed to handle. Leave a tube for a few minutes and it is plasma above and red cells below, so a pipette going in at a fixed depth draws something different every time. The established answers bought homogeneity with time: freeze and thaw the tubes, at the cost of a day, a freezer and condensation on every tube, or mix each sample on the deck and spend the run doing what a rocker does for nothing.

So I measured it: not whether blood separates, but how long a sample stays mixed after it comes off the rocker, and whether a pipette entering below the plasma layer reports the same concentration as the fully agitated sample. It does, and the window is wide enough to work inside. That turned an opinion into a specification: agitate the racks off the deck on an oversized blood rocker, de-cap, load, pipette wide-bore below the surface, and finish inside nine minutes, which is more time than the robot needs for two full racks.

Several colleagues had been through the vendor training before I took the deck on; what had not happened was a real method put on it and kept there. Two things made it work besides the programming: the tube adapters, so samples could go on the deck in the tubes they arrived in, and the order the other steps had to happen in. A full 96-well plate came out in 35 minutes with everything counted, equal or better than manual preparation on accuracy and precision, and the flow went into routine at 150 to 250 whole-blood samples a day.

Existing deck · no instrument purchased · posters at two IATDMCT congresses: Rotterdam 2015, a plasma multi-method; Kyoto 2017, this whole-blood flow, poster P25-6-7

The decks that followed, and four habits

That first method ran on a Hamilton STARlet the therapeutic drug monitoring laboratory already owned. On the strength of it the department went on to buy a Hamilton STAR M Plus for the doping control laboratory, ordered while I was at Hamilton Nordic, so I programmed that deck first as the vendor’s application specialist and then, back at Karolinska, as its user; then two STAR Ms for drugs of abuse, another STAR for therapeutic drug monitoring, and an Eppendorf epMotion 5075 for the pharmacogenetic PCR setups, TPMT and CYP genotyping among them. Five Hamilton decks and an Eppendorf across one department’s three laboratories, with me as the department’s main programmer from 2015.

Four habits came with every one of them. The worklist out of the LIMS and the tubes on the deck were matched barcode by barcode: a tube on the list but missing from the carrier was reported before the run started, and tubes loaded in any order were pipetted in the worklist’s order. The robot pipetted the calibrators, blanks and controls as well as the samples, so whatever controlled a sample controlled the curve and the QCs too. A manual version of every automated method was kept on par with it, equal precision and bias demonstrated rather than assumed, so that a deck going down changed the day’s pace and not its results. And a programmed method was judged by what came off the mass spectrometer, not by the deck reporting no errors: a plate-shaped heatmap of the internal standard response shows the systematic pipetting error that a step which ran without complaint hides, more internal standard in the later columns, say. That last check goes beyond a vendor’s handover, and I have done the handover.

Where the dried blood spot work starts

A dried blood spot is a cheap sample to collect and an awkward one to measure from, because how far blood spreads on the card depends on its hematocrit, so a punch of fixed diameter is not reliably a volume of fixed size. The method I am first author on determines that hematocrit from an image of the spot itself, which is one way of getting the volume back. The bench work belongs to these Karolinska years; the paper was published in 2023, after I had left for the laboratory in the next case. Dried blood spots come back twice more below, and the three papers are listed with their links on the about page.

ABC Labs: from covid testing to an accredited toxicology laboratory

• Situation

  ABC Labs in Stockholm, founded in 2020 as a high-volume pandemic testing operation, had to become an accredited drugs-of-abuse laboratory. The team of eight had been taken on for covid testing; one or two had mass spectrometry behind them.

• My role

  Chief laboratory technician for toxicology, hired in November 2021 to run the technical side and lead the team, with the QA function alongside. Employed 2021 to 2024.

• What changed

  A sample flow written first, then the mass spectrometer park installed and commissioned, LC-MS/MS methods developed and validated for urine, blood and oral fluid under a framework I wrote, QC, external quality assessment and measurement uncertainty set up, procedures and chain of custody written, staff trained.

• Outcome

  The scope, ISO 15189 for medical testing and EWDTS for workplace drug testing, was accredited in October 2023, 22 months from my first day. Accreditation is won by the laboratory and QA together.

• Evidence

  Swedac 10532 on the public register; a letter of support from the Medical Director, quoted below; two of the methods published and peer reviewed.

• Related service

  Laboratory commissioning and ISO 15189 and quality systems.

• 8People in the team, most of them new to mass spectrometry
• 22Months from covid testing to an accredited toxicology scope
• 3External quality-assessment schemes, reviewed against the scope

1. Nov 2021Hired into a covid testing laboratory
2. ↓Requisitions, sample reception, registration and the flow between them
3. ↓The mass spectrometer park, installed and commissioned
4. ↓LC–MS/MS method development and validation: urine, blood, saliva
5. ↓Quality control, external quality assessment, measurement uncertainty
6. ↓SOPs, chain of custody, training, audit preparation, CAPA
7. Oct 2023ISO 15189 and EWDTS scope accredited
8. ↓Methods and procedures developed further, the LIMS built out with the developers, staff trained, the move to Forskaren planned

My remit was the technical operation, end to end: the LIMS architecture and the sample flow it had to carry, requisitions, sample reception and registration, the analytical workflow, the instrument park and its commissioning, method development, evaluation, validation and measurement uncertainty, internal quality control and the external schemes it was checked against, SOPs and the validation framework behind them, reference material procurement, and the clinical chemistry that sits beside the mass spectrometry: creatinine in urine and amylase in oral fluid, because a drugs-of-abuse result means little until the specimen can be shown to be what it claimed to be. Method development was shared work, with a colleague hired for it and staff trained to carry parts of it; what I owned was the validation plans and reports, the data analysis between them, the risk assessments, and putting each method into production.

The part that took longest was not the instrument. It was teaching people to think in mass spectrometry and in drugs of abuse: what a ratio drifting out of tolerance means, why a clean-looking peak is not a positive, and when the honest answer to a customer is that the sample cannot be reported. Two of the methods developed during the build were dried blood spot methods, drugs of abuse and phosphatidylethanol, and the PEth method was the first one this laboratory took into routine. Both are published.

Accreditation was a milestone, not the end of the work. What followed was more of the same: procedures and methods developed further, the LIMS built out with the developers, new staff trained, the move to Forskaren planned. The register publishes the scope as it stands, not its history, so the October 2023 date is from my own account rather than from Swedac.

Miguel was hired in November 2021 to transit the ABC laboratory from molecular biology and covid testing to LC-MS and toxicology testing. He was responsible to set up the machine park of mass spectrometers, conduct validations, write standard operating procedures and teach technicians to run LC-MS.

Prof. Ola Winqvist, MD PhD · Medical Director, ABC Labs · letter of support, Stockholm, 18 November 2024

Employed 2021 – 2024 · eight-person team · ISO 15189 & EWDTS · Swedac 10532

The quality system

Built properly, a quality system is a set of decisions taken early enough to shape the workflow, so that the evidence an assessor asks for is a by-product of running the laboratory rather than a folder assembled for the visit. I ran the technical side of it, with QA alongside. Four parts of it are worth showing.

Document control, in a laboratory with no binders

ABC Labs kept no binders. The master of every controlled document lived in the laboratory’s own quality management system, a cloud system anybody entitled to read one could reach from any company device. Where the work needed a copy in front of it the bench had a printed one, issued and withdrawn like any other controlled copy. What did not exist was the uncontrolled version: the private printout in a drawer, the binder kept because the shelf is nearer than the system. The laboratory had been set up that way before I was hired; I worked in it and extended it.

Writing and issuing a procedure is not cheap and should not be: an author, a reviewer, an approver who is not the author, and QA. What a searchable system removes is the other cost, finding the current version and keeping every issued copy in step with the master. That is how I have worked since, and it is not a condition of working with me: your laboratory has its own document system, templates and version scheme, and I work inside them.

A validation framework, and what each kind of method has to pass

I wrote the guideline the laboratory’s method validations are run against. It defines twelve kinds of study, and twelve is the size of that set, not a count of validations performed.

• Bias and precision

  How close to the true value, and how close to each other on repeat. The pair every other line here is read against.

• Specificity

  Whether the signal being measured is the analyte, or something else that happens to arrive at the same moment.

• Precision at the decision point

  Precision measured at the cut-off, because that is the concentration where imprecision changes the answer rather than only the number.

• Calibration model

  Which curve, over which range, and whether it still holds its shape at both ends of that range.

• Carry-over

  What a high sample leaves behind for the one injected after it, and the level at which that stops mattering.

• Interference

  What else present in the sample can move the result, including the things a laboratory only meets in real specimens.

• Matrix effect and recovery

  What the sample itself does to the signal, and how much of the analyte survives the preparation.

• Limit of detection

  The concentration below which the method is not saying anything.

• Lower limit of quantification

  The lowest concentration the method is allowed to put a number on, which is not the same as the lowest it can see.

• Stability

  The analyte in the sample, in the extract, on the autosampler for as long as the workflow takes, and in storage afterwards.

• Accuracy and comparability

  Against reference material, and against what other laboratories get for the same sample.

• Agreement and average bias

  Against the method already in production, when a new one is replacing it.

The guideline maps each type of method to the subset it has to pass: a qualitative screening method and a quantitative confirmation method are not held to the same list. Two rules in it do most of the work. Nothing is run before it is published: the validation plan and the method procedure both go into the document system before the first experiment. And the roles are named: an owner who writes it, a reviewer, and an approver who is not the same person. It reached a second version after a laboratory non-conformity, as the custody procedure did. Validations under a framework like this are what method validation offers.

Control that is reviewed, not only recorded

Internal quality control across the scope: controls at the cut-offs that matter, limits set from the laboratory’s own running-in data, in-house manufacture of calibrators and controls with the ledgers behind it, rules for releasing a result and for handling a B sample. External quality assessment: the laboratory was enrolled in three schemes covering its reported scope, in both the drugs-of-abuse and the alcohol-marker directions, and what came back was reviewed against the methods. Measurement uncertainty estimated from the laboratory’s own precision and bias data, so that a result near a cut-off can be defended with a number.

Commissioning the instruments

When a new LC–MS/MS system was installed, the manufacturer’s engineer commissioned it against the manufacturer’s specification and it passed. That is the vendor’s obligation, and it is not the laboratory’s evidence. I wrote the laboratory’s own commissioning report on the same instrument before a result went to a customer: a repeatability benchmark on representative analytes in the room it stood in, the same benchmark repeated after a hardware change some weeks later, and a method comparison against the system already in production, thirty samples across a spread of positive and negative analytes, reported as bias with upper and lower limits of agreement. The question commissioning answers is whether the new instrument’s results agree with the ones the laboratory has already reported, and the supplier is not in a position to answer it. A relocation is a change like any other: when the clinical chemistry analyzer was moved, the creatinine assay on it was re-verified.

Before the instrument arrives

The most expensive question about an instrument is asked before it is bought: will it fit, and can it still be serviced once it is in? Before a purchase order is signed, I model the whole bench, instrument, sample organizer, LC stack, table and the clearance behind it, and issue a dimensioned drawing facilities can work from.

1. 1Sample organizer
2. 2LC stack
3. 3Xevo TQ-XS
4. 4Workstation
5. 5UPLC table
6. 6IonBench
7. 7Solvent bottles

I have taken the same instrument suite through a building move twice this way: out of Waters’ facility in Solna into Biomedicum, and then on to Forskaren. At Forskaren I modeled the rooms themselves and not only the instruments going into them, with the architect leads and against the manufacturers’ site preparation guides, so that what was drawn was what would pass review: the turn radius a wheelchair needs, the emergency showers, the outlets. The footprint is the easy part. The rest is gas lines, a nitrogen generator of the right purity, sound proofing, extract ventilation, cooling load and power, and where each of them has to land. Any one of those, found late, is a change order.

From the crate to routine work

An instrument that has passed its acceptance test is still not doing the laboratory’s work. In between sits a series of small decisions, and I have been on both sides of them: the laboratory taking delivery, and the vendor’s application specialist doing the handover. That gap is what laboratory commissioning offers as a service.

Site preparation, and an acceptance test that is yours

Site preparation for Hamilton liquid handlers across several models, for Eppendorf pipetting robots, and for the Waters Xevo LC–MS/MS benches drawn above. Then site acceptance protocols for the same systems: written tests the instrument has to pass where it stands, on the work it will be doing, with your matrix, your volumes, your carry-over limits and a pass criterion agreed before the run, rather than the manufacturer’s specification a second time.

Methods on the deck, including PCR setup

Hamilton decks for laboratories across the Nordics, programmed in 2016 as Hamilton Nordic’s application specialist, and Eppendorf epMotion robots for PCR setup, the reaction plate built by the robot instead of by hand at the step where a molecular laboratory is most exposed to contamination and to a plate laid out wrong. The deck layout is as much of the work as the programming: which position holds what, and what the method does when a tip is missing or a rack is loaded a hole out.

Worklists, and the instrument park behind them

The other half of a method is the worklist. At Karolinska and at ABC Labs it ran as one chain: out of the LIMS, through the liquid handler or the person filling the plate, onto the mass spectrometers, with nobody re-keying it at any step. The minutes saved at a keyboard were never the case for building it. The case is the error it closes off: a batch quietly acquiring the right sample under the wrong identity, cheap to design out beforehand and expensive to explain afterwards.

The instruments span both mass spectrometry houses, the clinical chemistry bench and the molecular side: Waters, on the ACQUITY and Xevo systems above, and Thermo Fisher; Olympus and Beckman Coulter analyzers, Thermo’s Indiko and Abbott’s Alinity on the chemistry side; Applied Biosystems StepOne for qPCR. No one laboratory ran all of these; this is the span across fifteen years.

Sample identity, LIMS requirements and controlled working records

• Situation

  The same ABC Labs build. A toxicology laboratory has to carry one sample identity from the requisition to the retained sample, across a customer ordering portal, a sample registration system, the LIMS and the instrument data systems, under chain-of-custody rules for workplace drug testing, and it has to be able to answer years later which lot, which operator and which instrument a result came from.

• My role

  I wrote the flow document and the systems chart, specified what the LIMS had to do and tested what the developers delivered, wrote the chain-of-custody procedure, and built the working ledgers the laboratory ran on until the LIMS could carry them.

• What changed

  One controlled document described the operation in nine steps across five sample streams; registration enforced the A and B tube rule; validated spreadsheets held the ledgers and then became the specification the developers built from.

• Outcome

  A laboratory that could show, for any result, who did what, when, to which sample and against which lot, and a LIMS built from rules already in daily use.

• Evidence

  The flow document and the chain-of-custody procedure are controlled documents in the laboratory’s quality system, which I cannot publish; the structure is described here. The accredited scope above was assessed against them.

• Related service

  LIMS requirements and laboratory workflow.

The first thing I wrote at ABC Labs was not a method. It was the flow: one controlled document describing the whole operation end to end, in nine steps, across five sample streams, drugs of abuse in urine and in oral fluid, the workplace-testing versions of both, and phosphatidylethanol in dried blood spots. Each stream branches differently, and writing them side by side is how you see which steps the five have to keep in common.

• Requisition

  What is being asked for, by whom, on whose authority, and what has to travel with the sample for the laboratory to be allowed to answer.

• Reception and registration

  Condition, seal, temperature, identity: what is accepted, queried and refused, decided at the door. Then the laboratory number, the position, and the checks the system performs rather than the person. This is where the physical sample and its record become the same object.

• Preparation

  Aliquoting, extraction, the screening and verification portions taken at the same time from the same tube, so a confirmation never depends on going back to a sample that has since been opened.

• Analysis

  Worklist, calibrators, controls, the run.

• Technical evaluation

  The run judged before the result exists: did the batch perform, are the controls where they should be.

• Technical approval

  A separate step from the evaluation, deliberately: the evaluation is the judgment, the approval is the named act of accepting it, so the record can say who decided and when.

• Medical approval and release

  The result judged before it leaves: is it consistent, is it reportable, does it need a comment.

• Result reporting

  Back to whoever asked, by the route agreed with them, answering the question the requisition was allowed to ask.

• Post-analytical handling

  Storage, at what temperature, for how long, with the ledger that records it; then disposal, also a controlled step.

The systems chart, drawn alongside the sample flow

The same document carries a second diagram: which system hands the sample to which. A customer ordering portal, a sample registration system, the LIMS and the instrument data systems, four pieces of software that between them have to carry one identity from the requisition to the report without a person retyping it. A sample flow without the systems chart produces a laboratory that works on paper; a systems chart without the sample flow produces software that models a laboratory nobody works in. My job was the translation between them: taking what the bench needs and putting it in a form a developer can build, then testing what came back against the way the work runs.

Chain of custody

Workplace drug testing raises the standard for every step, because the result may be contested by the person it is about. I wrote the laboratory’s chain-of-custody procedure against the European Workplace Drug Testing Society’s guidelines, and it is mostly a set of decisions about physical handling. The sample arrives as an A tube and a B tube under a single barcode. Registration prompts for the second scan and refuses a pair that does not match, so a mismatched set cannot be booked in. A and B are racked into separate, identically numbered racks, the same position in each, so finding the B for a given A never involves a search. The B racks go to refrigeration untouched; once a positive is verified, the A and its B move to a freezer under a long-term storage ledger, with retention agreed with the customer, and the A is held longer when a B has gone out to a reference laboratory. Anything that arrives in a condition the procedure does not allow goes into a deviations rack rather than into the run. Positive samples carry per-sample custody forms countersigned by a second technician, and a batch-level form for the set.

The version of that procedure in use is not the first one. It is the second, rewritten in response to a non-conformity found in the laboratory. A document with a revision history is evidence that a quality system is doing something, and it is hard to arrange after the fact.

Is the specimen what it says it is

A specimen that has been diluted, substituted or adulterated will still run cleanly on a mass spectrometer. So the toxicology workflow needed a clinical chemistry arm beside it: creatinine in urine and amylase in oral fluid on a clinical chemistry analyzer, to say whether the sample was plausibly what the requisition claimed. I set those up on the laboratory’s Indiko, and when two Abbott Alinity analyzers arrived later I transferred the methods across with instrument comparisons.

The ledgers, and what they have to answer

Four kinds of record carried the laboratory’s day-to-day accounting: controlled substances, the reagents it manufactured itself, equipment and calibration, and the instruments. What each of them gets asked, months or years later, has the same shape: what was it, who handled it, when, and against which lot.

Substances, reagents, and the ledger behind them

A toxicology laboratory procures and destroys controlled substances as a matter of routine, and has to account for all of it. I set up the substance inventory and the procedures around it: how a controlled substance is procured, how it is accounted for while it is in the building, and how it is disposed of. The inventory is reported to the Swedish Medical Products Agency, and accounting for every gram is a statutory obligation.

What the laboratory makes itself is held to the same discipline. Reagent ledgers for in-house calibrators, quality controls and mobile phases carry lot numbers, dates and a signature line for whoever prepared it, and the label is printed from the record itself, never written out by hand, so every bottle on the bench carries the right hazard pictograms. The mobile phase ledger also reads forward: it estimates consumption from use, so solvent is ordered while there is still time. A calibrator or a control the laboratory manufactures for its own use is, under the EU regulation on in vitro diagnostic devices, an in-house device, and the laboratory’s self-manufacturing was reported to the regulator on that basis; the ledger entry behind each batch is part of what was declared.

What is due, what changed, and where it is right now

Pipettes migrate: one is borrowed for an afternoon and is still on the wrong bench a year later, while the calibration schedule has it where it started. So the equipment ledger tracks position as well as date, and answers the dull questions on demand: what falls due this month, what is overdue, and which pipette is at which bench today. Each calibration carries the measured result, not a tick, and the name of whoever performed it. Pipettes go out to a calibration provider on schedule and are checked gravimetrically in house in between, against weights with calibration certificates of their own that trace back to a national standard, and those certificates hang off the same record.

The instrument records go down to the module. Each one carries its own part number and serial number, sample manager, column manager, solvent manager, ion source, so a service call opens with which module is misbehaving, and a swapped part is dated, so a method’s drift a month later can be laid alongside the day the ion source was changed. Every system carries a status, and the status does the work: in commissioning, nothing off it goes to a customer; in production, it does. Beside the status sit the links: the commissioning report, the acceptance protocol, the service visits, the maintenance log and the qualification documents, each one click from the record it belongs to.

Every field is validated where it is entered, which is what decides whether any of it is still true in two years. Dates that have to be dates. A status picked from a list rather than typed. A serial number that has to have the shape of one. A due date calculated from the last calibration and the interval instead of typed in by hand. A ledger that accepts anything is a description of the laboratory; one that refuses a bad entry at the keyboard is a record of it.

Where the documents lived

Two kinds of document, kept in two places on purpose: the controlled set, standard operating procedures, validation plans and the rest of that family, and the working record people reach for mid-shift. Both had to be findable. Only one had to be controlled.

Procedures written where people will look for them

Standard operating procedures were written into the laboratory’s own quality management system, through its review and approval route, rather than as documents in my format that somebody has to transcribe afterwards. Those systems are still the ones those laboratories run on, and a procedure that sits inside one keeps working after the person who wrote it has gone.

What the ledgers were built in

None of it needed a new system. The laboratory already had a cloud office suite, so the ledgers were built there, in spreadsheets, with no procurement and no waiting, and from the day they existed they were on every bench, every office machine and every phone. Built in a spreadsheet, but not run like one: each sheet was validated before anyone worked from it, versioned, and carried a change log; the parts not meant to be edited were locked. It was treated as critical software for as long as it was in use.

It was always meant to be temporary. The sheets held the ground until the developers could build each feature into the LIMS, and when they got there they were not working from a wish list. They had the sheet: every field, every rule, every list, in daily use and already agreed by the people who would live with it. The stopgap turned out to have been the specification. Meeting notes ran on the same principle, with a fixed shape, open to the whole company, signed by everyone they concerned and locked once signed. The controlled documents did not live there: procedures and the rest of that set stayed in the quality management system described above, and the substance inventory, a working ledger, was held to the same rules as the others, validated at entry, versioned, change-logged, its formulas locked.