Utilization of the Tyndall Effect for Enhanced Visual Detection of Particles in Compatibility Testing of Intravenous Fluids: Validity and Reliability

Particle Analysis by Light Obscuration and Turbidimetric Measurements

The results from the light obscuration analysis and turbidimetric measurements are summarized in Table I. The particles counted in each sample were divided into size classes, where the lower class was 0.5–2 μm and the upper class was particles above 25 μm, the upper limit of the Pharmacopeia specification. As the samples could not be diluted because of the risk of dissolving precipitates, the particle content in some of the samples exceeded the recommended concentration range of the detector (9000 particles/mL), meaning that these counts should only be taken as a rough estimate.

All samples, except the controls (sample 1–4), contained a considerable amount of small particles (0.5–2 μm). Even though the smallest particles may perhaps not be the most dangerous ones, it is alerting because microprecipitate may, depending on the specific kinetics, grow (e.g., Ostwald ripening) or aggregate with time. The formation of microprecipitate does not necessarily progress in an identical manner each time, for example, the rate or time of onset can be affected by surrounding factors such as temperature. This emphasizes the need to study the compatibility over time and to have proper controls for comparison to be able to detect changes. In the case of Y-site infusion, the typical contact time between drug and TPN in the line is 15 to 60 min. It is also recommended to analyze the samples after mixing at a later time point, for example, after 4 h (25).

Samples with precipitated calcium phosphate (samples 5–10) showed a gradual increase in particle number with increasing pH value, as expected (Table I). The diluted polystyrene samples (samples 11–15) were found to contain the theoretical particle concentration of the particle size in question (marked with grey background in Table I) with only small deviations, whereas most of the high concentration samples (samples 16–20) generally indicated particle contents somewhat lower than the theoretical particle content. The latter might be due to aggregation of the particles. The high concentration samples were not diluted, which implies a high particle density and may contribute to aggregation. The particle standards are designed for single-use validation of particle counters' precision, and transferring the particle suspension to glass tubes for repeated examination and storage were outside the recommendations and might have influenced the quality. The resuspension method prior to use did not seem to break aggregates satisfactory, and therefore some inaccuracy was found in the measured particle content of undiluted samples. Table I also shows that it was not possible to avoid the presence of some particles both above and below the desired particle size for all polystyrene samples. This is consistent with the terms of the manufacturers' certificate of analysis, for example, for the 25 μm standard the certified mean diameter is 24.61 ± 0.22 μm and the approximate concentration is 2000 particles/ mL [(±10%) ≥ 15 μm]. Nevertheless, these standards were readily available, and because of their given concentration it was easy to dilute them to other concentrations. The particle-free controls (samples 1–4) showed low particle counts, although they contained a few particles, demonstrating that total freedom of particles is not an achievable goal.

The turbidimetric measurements confirmed a low turbidity of the particle-free controls. The precipitated samples showed a gradually increase in turbidity with increasing pH value, starting with a turbidity value similar to the controls (sample 5). The low-concentration polystyrene samples had overall very low turbidity values. The more concentrated samples showed higher turbidity values, but not as high as one might expect, considering that the particles were detectable also without utilizing a Tyndall beam. This indicates that the polystyrene particles transmit a lot of light. It has been suggested that a change in turbidity of 0.5 NTU as compared to the turbidity of the fluids before mixing may be defined as incompatibility (42). [FNU is equivalent to NTU in low regions, i.e., up to 40 NTU (43)]. Applying this limit, samples 6–10 and 17–20 would be regarded as incompatible or different as compared to the original, for example, the controls. Although it can be useful to define a limit like this, it is still important to evaluate and compare the result with the original solutions (controls) and look for overall changes.

Classification and Appearance of Samples by Visual Examination

The classification of samples, based on the inspectors' evaluations, is summarized in the rightmost side of Table I for both Tyndall beams. The six precipitated samples (samples 5–10) were all located in the reject zone (rejected by ≥70%) for examinations using the pocket laser pointer, whereas only the four samples with the highest pH (samples 7–10), that is, containing the highest amount of precipitate, were located in the reject zone using the focused desk lamp. The remaining two samples were found in the grey zone (rejected by <70% to >30%). Consulting the results from the particle analysis and turbidity measurements, this suggests that the laser pointer was quite sensitive in the detection of fine particles (<2 μm). Interestingly, more than 60% of the inspectors also reported identification of a Tyndall effect in the TPN_aq controls (samples 1 and 2) with the laser pointer. Comparing the rejection rates with the results from the light obscuration and turbidimetric measurements, this finding may be regarded as false-positive. Also, in the wider light beam of the focused desk lamp, both TPN_aq controls were classified in the grey zone, as 40–45% of the inspectors found the TPN_aq controls to be different compared to the Milli-Q controls. The inspectors described the TPN_aq controls as having a very weak bluish to white hint/opalescence compared to the clear, transparent appearance of Milli-Q water. Also, Trissel and co-workers commented that the PN solutions investigated with their Tyndall beam appeared as hazy liquids (24). The Tyndall effect observed in TPN_aq samples seems to be an inherent property of the parenteral nutrient admixture. For instance, it is known that solutions containing fluorescent compounds can display a shaft of light similar to the Tyndall effect, when illuminated, and amino acids, which are part of TPN, can have this property (28). The authors have also experienced fluorescence with other drug solutions. This complicates the visual evaluation. It might also be other interactions between the components of the TPN admixture that contributes to the hazy appearance. The fact that the clear TPN_aq controls were falsely rejected in the current study may be a warning that the laser pointer in particular may cause problems with too high frequency of false rejection in a compatibility test set-up involving complex mixtures. However, the Milli-Q controls were, as expected, found in the accept zone with both light sources. It is therefore plausible that the method is better suited for compatibility testing of simpler drug solutions without an inherent background noise, as described above for the TPN_aq controls.

The appearance of the polystyrene particles was described as different from the calcium phosphate precipitate in both light sources. The polystyrene particles were more translucent and were sometimes mistaken for gas bubbles. The large polystyrene microspheres in low concentrations seemed to light up when illuminated with the laser pointer, but they were too few to form a continuous line. Because the laser beam was narrower than the width of the test tube and therefore less likely to hit and illuminate the particles, it was more difficult to detect particles at low concentrations. The wider beam of the focused desk lamp seemed somewhat better at identifying low numbers of larger particles. All high-concentration polystyrene samples (samples 16–20) were classified in the reject zone with both light sources, all with very high rejection rates. The high concentration of particulate matter (undiluted samples) was easily identified regardless of the actual size of the particles; they could also be recognized in normal light with the naked eye. Therefore, the diluted samples (samples 11–15) were more interesting. For these samples a higher rejection rate was found using the focused desk lamp compared to the laser pointer. It is interesting to notice that the sample with 2 μm microspheres 100/mL (sample 11) was accepted (30% rejection probability) with the laser pointer and found in the grey zone (35% rejection probability) with the focused desk lamp, whereas the 5 μm microspheres 50/mL was rejected by 50–55% of the inspectors. The larger microspheres (>5 μm) were mostly rejected by more than 50% of the inspectors (grey zone), regardless that the particle concentration was much lower. This suggests that the lower particle size limit for visual detection using a Tyndall beam is in the area around 5 μm, as also indicated by Veggeland and Brandl (22). Based on the findings, we propose that such a detection limit is primarily interesting for samples with a low number of particles—for instance, few precipitated crystals or other sources of particle contamination than incompatibility. In a sample with a high number of small particles, the use of a Tyndall beam will allow detection of even smaller particles, as the high number of particles scatters light sufficiently to be detected with the help of the Tyndall effect. This even applies to particles in the nanometer size range (e.g., samples 5 and 6).

The samples containing particles corresponding to the Pharmacopeia limits for sub-visual particles in LVPs, that is, 10 μm 25/mL (sample 13) and 25 μm 3/mL (sample 14), were rejected by 55–60% of the inspectors in the current study. This means that the inspectors were highly disagreeing on the evaluation, and that most of them would reject the sample as containing particles. This suggests that both sources of Tyndall beam could identify particles in this size and concentration range, but neither was robust enough to give consistent results in particle detection on the boarder of what is acceptable for sub-visual particles. The pharmacopeia limits are the only formal, numerical requirements regarding particle content of parenteral fluids, and are therefore interesting as comparison. It was easier to standardize samples with a particle content corresponding to the limits for LVPs because the limits are given per milliliter. SVPs have limits of total numbers per container. This results in different amounts of particles per milliliter. A SVP of 10 mL containing 6000 particles in total ≥10 μm would give a particle concentration of 600/mL ≥10 μm. Based on our findings, this would have been easier to detect than 25 particles/mL ≥10 μm. Therefore, the visual examination method using Tyndall beam might be more suitable for examination of SVPs as compared to LVPs. Nevertheless, it should be kept in mind that an increase in the number of small particles (sizes below of the Pharmacopoeia limits of 10 microns) is also relevant with respect to compatibility.

Validity and Reliability of the Examination Method

To further evaluate each of the light sources and compare its usability to the other, validity and reliability parameters were calculated (Table II and III, respectively). Generally, only small differences were found between the two light sources, and none of the differences were statistically significant. The specificity was slightly higher for the broad frequency white light of the focused desk lamp compared to the narrow frequency band of the red pocket laser pointer, suggesting that the focused desk lamp might be better at evaluating the negative samples correctly; that is, the negative samples were less frequently falsely rejected. Also, the LR+ was slightly higher for the focused desk lamp compared to the laser (Table II), again supporting the hypothesis that there is a lower possibility of a particle-free sample being rejected using this light source. The sensitivity was almost the same with both light sources. The LR+s were above 1, meaning that a sample containing particles is more likely to be rejected than a particle-free sample. To further understand the meaning of the size of the likelihood ratios, we conferred Jaeschke and colleagues' user guide (44): Likelihood ratios in the range 2 to 5, as estimated for both light sources, only give a small, but sometimes important, increase in the probability of the sample containing particles compared to the pre-test probability (i.e., the probability that the sample contains particles before the examination is performed). This can be interpreted as an indication that the method is only sometimes useful for its purpose. Higher LR+ values increases the power of the change, and LR+ values above 10 would be indicative of large, conclusive changes. The LR−s were estimated to be below 1 (Table II), meaning that a particle-free sample is more likely to be accepted than a particle containing sample. According to Jaeschke et al., negative likelihood ratios of 0.2–0.5 indicate only a small decrease in the probability of the sample containing particles compared to pre-test probability (44).

Regardless of the light source used, the inspectors expressed uncertainty during evaluation, most frequently with regard to the samples containing low concentrations of particles. Interfering factors, such as scratches on the glass walls and micro-bubbles in the samples, may also have confused the inspectors. More training and experienced inspectors could affect the outcome of this type of study.

Kappa statistics are frequently used to assess inter-rater agreement, but it is associated with weaknesses described as the kappa paradox (40). Gwet's AC1 was developed to overcome these challenges (40). Therefore, both inter-rater agreement coefficients were used in the analysis of reliability. Both the Fleiss' kappa and Gwet's AC1 were found to be somewhat higher for the laser pointer as compared to the focused desk lamp (Table III)—however, not on a statistically significant level. The 95% confidence intervals were rather wide. Based on the Landis and Koch scale, the AC1 values were interpreted as a moderate agreement between the inspectors using both light sources, whereas the Fleiss' kappa value indicated fair agreement for the focused desk lamp and moderate agreement for the laser (41). A fair to moderate agreement between the inspectors is not good enough to recommend this method alone for compatibility testing, as this means that the result would strongly depend on the inspector conducting the examination. Used in compatibility testing this could represent a risk of not detecting the incompatibility of two components that are in fact incompatible. However, scrutinizing the inter-rater agreement coefficients obtained for only the precipitated samples (more realistic samples) with the laser pointer, the inter-rater agreement factors increased; Gwet's AC1 of 0.83 (95% CI: 0.52–1). Therefore, it seems that the evaluations of the polystyrene samples caused the most disagreements and were responsible for pulling the inter-rater agreement down; the method could be better than the overall results imply. Even though polystyrene standard particles were easily available in the desired particle sizes and concentrations, they may appear different in examinations as compared to precipitated particles formed as a result of incompatibility.

In a patient safety perspective a false accept would have worse consequences than a false reject, and in the current study the false accepts were more associated with the (perhaps less relevant) low-concentration polystyrene samples. On the other hand, when using a test that gives many false rejects one might ask if the inspectors are rejecting the true-positive samples because they actually see particles or if they are guided by other features (e.g., background noise). If this feature was absent, would the true-positives still be rejected? For example, would precipitated samples without the background noise of TPN be rejected if they contained the same amount of particles? If no, the method could be less sensitive when applied to other sample sets.

Of note, there were more positive (particle-containing) than negative samples (particle-free) in the test set that the inspectors were examining, which may have caused imbalances in the results. A wide range of particle sizes as well as concentrations were investigated; nevertheless, more or different samples might give a different result. In future studies it might be interesting to include precipitates with a different appearance than calcium phosphate as well as more simple formulations than TPN_aq. Because the examinations had to be conducted over several days in order to allow as many as 20 inspectors evaluating 20 samples, the samples were not identical at the time point for each of the visual examinations. The precipitated samples were freshly prepared each day, and due to the dynamics of precipitation the particle content and particle size in the samples could differ depending on whether they were examined early in the day or later. This was accounted for by analyzing samples before and after examinations, and some differences were observed (data not shown). Scrutinizing the evaluations given by the inspectors in detail, no clear tendency could be identified that corresponded to different evaluation of the samples depending on the time of the day they were examined (data not shown).

Other light sources, such as for instance cold fiber optic, might have some advantages over the ones investigated in this study. Possible advantages might be less heat generation and brighter light. The authors have performed some preliminary studies with a cold fiber optic light source and it seemed to provide a somewhat brighter picture of the samples, but the drawback was that the interfering factors were enhanced also. In other words, the inspector might still be confused and experience similar difficulties drawing a conclusion using more advanced light sources as well. In terms of the red pocket laser it can be argued that lasers with shorter wavelength, like blue or green lasers, and hence higher intensity light scattering would be more sensitive than the one used in the current study. The reason for choosing a red laser was that it was inexpensive and readily available, but also because such a laser had been used in another study discussing the Tyndall effect and particles in parenterals (22). Moreover, in preliminary studies we found that the green laser enhanced not only the particles of the samples, but also the background in particle-free samples; even in Milli-Q water a weak green line was detected. It was therefore regarded as plausible that using a shorter wavelength laser would increase the number of false-positives. This illustrates again the challenges of selecting the appropriate light source, but also the problems encountered when relying only on human vision to distinguish between the truly “bad” and “good” samples. Comparing the two tested light sources, it was found that the sensitivity of the narrow-band red light was approximately the same as for the broad-frequency white light, whereas the specificity of the red light was slightly lower (Table II).

Another aspect worth discussing when it comes to the design of compatibility studies is the method used to mix the samples. In the current study the mixing is done in test tubes with fixed concentrations, which represents a static compatibility test. It could also be possible to design a dynamic setup in order to achieve a more realistic simulation of drug meeting TPN in the infusion line. In real Y-infusion, different concentration gradients, temperature gradients, turbulence, and other events may occur and influence the mixing. These are all factors that are not easily simulated in a simplified static test.

Overall Discussion

Defining a specific size limit for particles considered harmful is challenging because the full extent of harmful effects of particles is not known. It is also complicated to define a limit to which particles can be considered visible because the detection of particles is dependent on several aspects in addition to size. Even small “sub-visual” particles might be detectable if they are large in number due to their combined light scattering. It has been suggested that the lower limit of visibility for the naked eye in normal light is around 40 μm (45), which may be lowered by the use of the Tyndall effect, possibly to a particle size around 5 μm (22); some reports claim even down to 1 μm (46). Our findings suggest that the lower detection limit for particles might be close to 5 μm. Nevertheless, defining a detection limit is probably more interesting for samples containing single or few particles. In compatibility testing it is important to detect changes occurring as a consequence of mixing (i.e., the evolution of particle growth and haze), which might begin with tiny particles increasing in number and size over time. Rather than focusing on the size limit, the suitable method should be able to detect indicators of change in mixed samples as compared to un-mixed controls. This can be equally challenging because the changes originating from incompatibilities are not always reproducible, as illustrated by Parikh and colleagues (8). They found calcium phosphate precipitation in a parenteral nutrient bag 30 h after compounding. The particles were detected by light obscuration, but could not be seen by visual examination using a “high-intensity lamp”. When they repeated the exact same experiment, even higher numbers of particles were counted in the light obscuration analysis, and this time the precipitation was also detected by visual examination. The experiment was apparently reproduced, but some factors beyond their control influenced the result, and the conclusion was different.

In a hospital pharmacy setting where access to advanced analytical instruments is limited, simple and inexpensive method like visual examinations seems attractive. It has been proposed that the visual method may have some advantages over electronic methods in cases were particles adhere to the container walls or when the particles are not evenly distributed in a sample (24). In the current study, the low-concentration polystyrene samples seemed more easily detected by visual examination than by turbidimetric measurements because the particles were translucent and gave only small changes in turbidity. However, the turbidimeter was sensitive in detecting changes from clear controls to increasingly precipitated samples. Other reports emphasize the benefit of an objective method for the detection of sub-visual precipitates, such as light obscuration, over the use of exclusively visual methods (21). Trissel and co-workers also draw attention to the fact that particle formation and turbidity may have been overlooked in some of the earlier compatibility studies with PN admixtures, as they did not evaluate sub-visual physical phenomena (24). Also, recent studies base the detection of particles entirely on visual detection, for example, screening for calcium phosphate precipitates in neonatal PN mixtures (47), but they do suggest using in-line filters as standard procedure to avoid injecting particles of risk. Nevertheless, the current study suggests that the validity and the reliability of the visual test are not good enough to ensure safety of the patients based entirely on this test alone. This is supported by Driscoll et al. (48) as well as the safety alerts regarding deaths following injection of calcium phosphate precipitates (1). It should be emphasized that in-line filters cannot replace proper compatibility testing, but they may be used as a secondary safety precaution to protect the patient.

Based on the findings of the current study, large particles and pronounced precipitations would be detected in visual examination with the Tyndall beam, but when the particles were smaller in size and/or in lower concentrations the inspectors expressed a considerable degree of uncertainty and the inter-rater agreement decreased. Our results suggest that the method is not optimal as a single method for compatibility screening for the purpose of parallel infusion of PN and drugs. If precipitation is observed, the method can offer support to advise against parallel infusion. However, lack of detected particles/precipitates cannot be interpreted as clear evidence of no compatibility problem for the respective products. Thus, we conclude that the complex issue of compatibility should be based on careful evaluation of data founded on more than just one visual examination. No method alone is without weaknesses, instrumental methods being no exception. Therefore, a program of several methods evaluating different indicators of incompatibility and providing a broad basis for evaluation of compatibility issues is recommended. Visual examination may serve as one of the methods in a screening program (27, 30). In a critical situation where no documentation is available and a more extensive analysis, including instrumental methods, is not accessible, the visual examination method using a Tyndall beam to enhance the visibility of sub-visible particles may serve as a resource (20). Using an in-line filter is paramount as a secondary safety precaution to protect the patient (48).

To increase safety in clinical practice, research labs and the pharmaceutical industry should have a continuous focus on compatibility testing to provide quality-controlled documentation on compatibility.