evaluation of random blinded re-checking of afb slides under

Ó Journal of Krishna Institute of Medical Sciences University

JKIMSU, Vol. 2, No. 1, Jan-June 2013

ORIGINAL ARTICLE

Evaluation of Random Blinded Re-Checking of AFB Slides under RevisedNational Tuberculosis Control Programme in Solapur District

Swapnil Vishnu Lale*District Tuberculosis Officer, Civil Hospital, Solapur - 413002, (Maharashtra), India

Abstract:

Background: One of the important componentsof revised national tuberculosis controlprogramme is �Good quality diagnosis, prima-rily by sputum smear microscopy�. All effortsare made to ensure that the designated micros-copy centers function at optimal level. The pro-cess of �Random Blinded Re-Checking� (RBRC)of Acid Fast Bacillus slides is built in theprogramme. Objectives: To study the relation-ship of different types of errors detected inRBRC with respect to time, place and cost. Tostudy the stability and capability of the processof RBRC. Methods: Analysis of secondary dataof external quality assessment of Solapur dis-trict since January 2006 is supplemented bydirect implementation of the programme sinceApril 2011 till date. Data analysis is done usingstatistical software Minitab version 16.Results: Since January 2006 to May 2012;42191 slides were re-checked in 77 RBRC ses-sions at District Tuberculosis Center, Solapur.Different types of 69 errors were detected. On-site evaluation and panel testing did not showany discordance. Barshi and Mangalwedha Tu-berculosis Units (TU) showed significantlyhigher number of errors as compared to KarmalaTU. (P<0.002) Weighted Pareto Chart revealedthat the costliest form of errors is high falsenegatives and low false negatives. Conclusion:Detection of errors in RBRC sessions follows

Poisson distribution. The process of RBRC isfound to be in control and capable of achievingthe desired target of detection of errors.

Key words:

DPU- Defects per unit, HFN- High false nega-tive, RBRC- Random blinded re-checking

Introduction:

Revised national tuberculosis controlprogramme (RNTCP) was pilot tested from1993 to 1997. RNTCP was scaled up in a phasewise manner to cover the entire country byMarch 2006. It was implemented in Solapurdistrict from April 2002. Directly observedtreatment with short course chemotherapy(DOTS) is a core strategy of RNTCP. One ofthe important components of DOTS strategy is�Good quality diagnosis, primarily by sputumsmear microscopy.� Microscopy also facilitatescategorization of patients; monitor the re-sponse to the treatment. Quality assured smearmicroscopy laboratories are established for thispurpose. An effective quality assurance systemis built in the RNTCP programme. The formerconsists of 1) Internal quality control, 2) Ex-ternal quality assessment (EQA) and 3) Con-tinuous efforts for quality improvement oflaboratory services [1]. The system also pro-vides credibility of laboratory results and mo-tivation of the staff for further improvement inthe efficiency. This paper studies the external

32

ISSN 2231-4261



quality assessment aspect only.This paper attempts to describe the distributionof different types of errors detected in EQAprogramme of Solapur district with respect totime and place. Identification of poor perform-ing tuberculosis unit (TU) helps the programmemanager to pay more attention towards that par-ticular unit. Identification of low performancetime period may point towards certain deter-minants. The secondary data available at the dis-trict tuberculosis center (DTC) from January2006 till date is subjected to analysis byMinitab software and salient findings are pre-sented in this paper. The main objectives are tostudy the relationship of different types of er-rors detected in random blinded re-checking(RBRC) of AFB slides with respect to time,place and cost and to suggest suitable recom-mendations. To study the stability and capabil-ity of the process of RBRC.

Material and Methods:

The population of Solapur district is 43,15,527. Solapur city has a population of951118; while the population of Solapur ruralis 3364409 [2]. The latter is divided into sevenTuberculosis Units (TU). They are DTC Solapur,Akkalkot, Barshi, Pandharpur, Akluj,Mangalwedha, and Karmala. A total of 33 Des-ignated Microscopy centers (DMC) are func-tioning in these 7 TUs. One senior tuberculo-sis treatment supervisor (STS) and one seniortuberculosis laboratory supervisor (STLS)works in each TU. District Tuberculosis Cen-tre (DTC) is responsible for quality assuranceof smear microscopy for all these 33 DMCs.In this capacity, DTC performs External qual-ity Assurance (EQA). EQA has three compo-

nents. On site evaluation (OSE), b. Paneltesting, c. Random blinded re-checking of rou-tine slides.a. On-site evaluation- On-site evaluation is

conducted at least once a month by STLSat the DMC. The visit includes a compre-hensive assessment of laboratory safetyprocedures, condition of equipment, ad-equacy of supplies as well as technicalcomponents of AFB smear microscopy likepreparation, staining and reading of smears.

b. Panel testing- This method evaluates in-dividual performance in staining and read-ing and not all the laboratory activities.Panel testing is conducted by State tuber-culosis training and demonstration centre(STDC), Pune at DTC, Solapur.

c. Random Blinded Re-Checking (RBRC)of routine slides - This EQA method pro-vides reliable assurance that a district hasan efficient AFB microscopy laboratorynetwork supporting RNTCP. Blinded re-checking is a process of re-reading a sta-tistically valid sample of slides from a labo-ratory to assess whether that laboratory hasan acceptable level of performance.Sample is collected using Lot Quality As-surance Sampling (LQAS) method. LQASrequires smaller samples because it doesn�tattempt to construct a precise estimate ofpopulation parameters. RBRC is con-ducted every month [3].

The STLS selects from the laboratory registerthe sample slides for RBRC using LQASmethod. The laboratory technician (LT) pre-pares a list of the slide numbers that are se-lected by STLS along with results and enclosesin them a sealed envelope. LT arranges the

Swapnil Vishnu Lale

33



slides in a separate box supplied by DistrictTuberculosis Officer (DTO) and marks on thetop of the box as well as an envelope with thetitle: LQAS slides, Name of DMC, TU and themonth and year. STLS picks up the box and theenvelope and hands them over to DTO. DTOcodes and interchanges slide boxes amongSTLS, retaining the sealed envelope in his pos-session. Thus blinding of slides is ensured bythe DTO. STLS reads and records the resultsfor slides. Umpire reading will be done by an-other STLS selected by the DTO on the basisof merit to resolve the dispute. According toresults from the DMC and results from ran-dom blinded re-checking; different types oferrors emerge. The errors are classified in fivecategories

a. High false positive (HFP) - Here DMCresults shows high bacillary count butactually the slide is negative.

b. High false negative (HFN) - Here theresult from DMC is negative, but in re-checking high bacillary count is noted.

c. Low false positive (LFP) - ThoughDMC has reported scanty bacilli but inre-checking the slide is found to benegative.

d. Low false negative (LFN) - Here theDMC has labeled this slide as negativebut during rechecking scanty bacillarycount is observed.

e. Quantification error (QE) - Here thereis discordance between DMC and re-checking about the quantification.(Scanty and high bacillary count)

DTO gives feedback and corrective action tolaboratory technicians through the Medical of-ficer of the DMC. A variable follows a Pois-

son distribution if the following conditions are met.a. Data are counts of events. (Data must

contain non-negative integers with noupper bound)

b. All events are independentc. Average rate does not change over the

period of interestThe variable �number of errors� fulfills all thesethree criteria. Therefore assumption is madethat the variable �Total number of errors� fol-lows Poisson distribution during each sessionof RBRC during entire duration of observation.This distribution is described by one parameterLambda. This parameter equals mean and vari-ance.Capability analysis of the process of RBRCWorking definition of Process - A process is aunique combination of instruments, chemicals,methods and technicians involved in producinga measurable output; for example : RBRCprogramme. All processes have inherent statis-tical variability which can be evaluated by sta-tistical methods. Process capability is a mea-surable property of a process to the specifica-tion, expressed as process capability index oras a process performance index. The output ofthis measurement is usually illustrated by a his-togram and calculations that predict how manyparts will be produced out of specification. Twoparts of process capability are

1) Measure the variability of the output ofa process (Errors per RBRC session)-

a. U Chartb. Cumulative DPU

2) Compare that variability with a proposedspecification (or a target)

a. Poisson plotb. Histogram of distribution of DPU

Swapnil Vishnu Lale

34



c. Summary statistic tableThis paper analyses the secondary data collectedin the DTC since January 2006. Retrospectiveanalysis is supplemented by direct implemen-tation and supervision of the programme sinceApril 2011 till May 2012. While performingthe analysis of quality of any process; it is pru-dent to use popular, established and widely ap-plied quality assessment software. Thereforedata analysis is done using statistical softwareMinitab version 16.

Results:Table 1 shows the dates of proficiency paneltesting by STDC, Pune. The reports of STLSfrom all the TUs show acceptable to good per-formance. The analysis of on-site evaluationcheck lists from January 2006 to May 2012 didnot show any discordance.Since January 2006 till May 2012, a total of77 RBRC sessions were conducted in DTC,Solapur. 42,191 slides from 7 TUs and 33DMCs were re-checked. Different types of 69

Swapnil Vishnu Lale

35

Table 1 - Results of OSE and Panel testing

01-06-2005

2006

2007

2008

10-09-2009

13-07-2010

22-08-2011

18-06-2012

Date of testing DTC Akkalkot Barshi Pandharpur Akluj Mangalwedha Karmala

Acceptable

Good

Good

Good

Good

Good

Good

Good

Good

Acceptable

Acceptable

Good

Good

Good

Good

Good

Good

Good

Good

Good

Acceptable

Good

Good

Good

Good

Good

Good

Good

Good

Good

Good

Good

Good

Good

Good

Proficiency panel testing was not conducted during this period.

Analysis of on site evaluation check lists over the period from 2005 to 2012 did not show any discordance

Table 2 - Year wise distribution of different types of errors

2006

2007

2008

2009

2010

2011

2012

Total

Year

Data upto 31st May 2012

Total number of slides examined Total number of errorsHFP HFN LFP LFN QE

6140

6596

6033

6428

7440

6269

3285

42191

14

16

8

8

8

10

5

69

1

1

0

0

2

0

0

4

6

5

0

0

2

5

3

21

2

0

0

1

1

0

0

4

1

1

0

1

0

2

0

5

4

9

8

6

3

3

2

35



errors were encountered during the process.Year wise distribution of different types of er-rors are shown in Table 2. Maximum numberof errors were found in years 2006 and 2007.During this period proficiency panel testing ofSTLS was not conducted by STDC. (Refer toTable 1) Most common type of errors are quan-tification errors, followed by high false nega-tives

Before conducting statistical analysis, Minitaboffers to graphically explore the data and as-sess relationships among the variables. Figurenumber 1 shows the grouped histogram, whichdisplays the histograms for each tuberculosisunit on the same graph. X axis shows the totalnumber of errors and Y axis denotes the den-sity (frequency). Table embedded in the samefigure shows TU wise mean number of errors,

Swapnil Vishnu Lale

36

Fig. 1 - TU wise distribution of total number of errors

Fig. 2 - Scatter plot of total number of errors vs. years

N. B. Data up to31st December 2011



standard deviation, and total number of errorsin each TU. Each TU is depicted in differentcolour, explanation of which is provided in thelegend. This grouped frequency polygon showsthat DTC, Akkalkot and Akluj are similar in meannumber of errors and spread of errors. In con-trast, Barshi TU has more number of errors andmore spread out. Later in this paper, efforts aremade to detect the statistically significant dif-ferences among means using analysis of vari-ance.This data was obtained from January 2006 toMay 2012. But for year wise comparison dataup to the completed year 2011 is taken into con-sideration. As data spans over six calendar years,we suspect whether detection of errors changesover the period of time. To verify this suspi-cion and eliminate time period as a potentiallyimportant factor, the relation between the num-ber of errors over period to years is examined.Fig. 2 shows 7 different panels for each TU. Xaxis representing the years ranging from 2006to 2011. By custom, this is a predictor vari-

able. Y axis represents the total number of er-rors in each year. The blue lines are the regres-sion lines. The points on the scatter plot ex-hibit no apparent pattern in any of the seven TUs.The regression line for each centre is relativelyflat, suggesting that total number of errors are

Swapnil Vishnu Lale

37

not affected over the years.

One-way ANOVA: Total no of errorsversus TU

Descriptive statistics for each TU such as meannumber of errors, standard deviation and totalnumber of errors is provided in table embed-ded in Fig. 1. Barshi and Mangalwedha TU havemore number of errors and higher standard de-viations; while Karmala TU has lower meannumber of errors and smaller standard devia-tion. By graphical analysis it seems that the dif-ference in number of errors across differentTUs is statistically significant. To verify this,one-way ANOVA test was performed, whichtests the equality of two or more means cat-egorized by a single factor. Null hypothesis (H

0)

is occurrence of errors is independent of TUs.Table 3 shows that the value of F ratio is 3.98.The calculated value of F is greater than the tablevalue. P value is 0.002. This P value providessufficient evidence that the mean number oferrors is different for at least one TU from the

others; when α is 0.05. H0 is rejected and alter-

native hypothesis �Occurrence of errors is de-pendent on different TUs� is accepted. Tukey�smultiple comparison tests is done to test whichTU means are different.

Table 3 - One way ANOVA table

S=0.65, R - Squared = 29.52 %, R-Squared adjusted = 22.10 %, N.B. Data up to 31st December 2011

6

57

63

Degrees of Freedom

Between the TU

Error (within the TU)

Total

Source

10.11

24.13

34.23

Sum of squares

1.68

0.42

Mean sum of squares

3.98

F Ratio

0.002

P



In Fig. 3, Tukey�s method is used to create con-fidence intervals for all pair wise differencesbetween TU level means while controlling thespecified family error rate of 5 %. In the 95 %confidence intervals table, intervals of KarmalaTU did not overlap the interval of Barshi/Mangalwedha TU. Also the means of Barshi/Mangalwedha and Karmala do not share groupletters A and B. They are statistically different.Less number of errors are found in KarmalaTU while significantly more number of errorsare found in Barshi and Mangalwedha TU.

Weighted Pareto chart- Calculating the costof poor quality

Pareto chart is a special type of bar chart wherethe plotted values are arranged from largest tosmallest. Fig. 4 shows that quantification er-rors is most common type of error and it isfound in 35 instances. It is followed by highfalse negative (HFN) and low false negative(LFN) observed for 21 and5 times respectively.Least common errors are high false positives(HFP) and low false positives (LFP) found in 4instances each. Pareto chart is a basic qualitycontrol tool used to highlight most frequently

Swapnil Vishnu Lale

38

Fig. 3- Tukey�s multiple comparison tests



occurring errors. Pareto chart is named afterVilfredo Pareto and his principle of the �80/20 rule�. That is 20 % of people contain 80 %of wealth; or 20 % of errors cause 80 % ofloss.A weighted Pareto chart not only considers thefrequency of occurrence, but the importanceas well. A weighted Pareto chart can accountfor the loss caused by mislabelling a sputumpositive tuberculosis patient as sputum nega-tive or denying the anti-tubercular treatment.It may be the risk of spread of TB or likely deathof the patient. In this study data on the fre-quency of occurrence of error is collected. Thecost required for aversion of each DALY dueto tuberculosis is $13.3 to $103.5 [4]. So ar-

bitrary valuation in this range for each missedcase of tuberculosis is Rs. 4500/-. Same valueis used for each case of TB who is unnecessar-ily treated.Fig. 4 shows that most costly errors are HFNand LFN. Recommendations should be directedat reducing false negative errors.Poisson process - Poisson process describesthe number of times an error occurs in finiteobservation space. Here it describes the num-ber of errors detected in RBRC sessions. Table4 show the total number of errors detected inmonthly RBRC session. 69 errors are spreadout in 77 RBRC sessions during a time span ofsix and half years in purely random manner.

Swapnil Vishnu Lale

39

Fig. 4 A - Weighted Pareto chart Showing Frequency

The most frequently occurring error is quantification error (QE). Based just on this informa-tion, one may decide to develop an improvement project around reducing QE.



Table 4 shows that in this RBRC process lambdais equal to 0.8961.

Capability analysis of the process of RBRC

Fig. 5 shows five components of capabilityanalysis of process of RBRC.

1) U chart - The U chart is located in theupper left corner of the fig. 5, Processcapability analysis. U chart is used todetermine whether the number of errorsper RBRC session is in control. Plot-ted points represent the number of er-rors per session of RBRC. Center line(green) shows the average number oferrors per RBRC. Control limits (red)

are located 3 δ above and below the cen-ter line, and provide a visual means forassessing when the process is out ofcontrol. In this study there are 4 pointslocated outside the control limits.

2) Cumulative DPU - The cumulative DPU(defects per unit of measurement) graphhelps to determine whether the samplesize is adequte to have a stable estimateof the DPU. This graph is located in thelower left corner of the fig. 5. Thisgraph shows that the DPU stabilizes af-ter several samples (at 55th RBRC ses-sion onwards) at 0.89. This appears as aflattening of the plotted line. This sta-

Swapnil Vishnu Lale

40

Fig. 4B- Weighted Pareto Chart showing cost of errors:

The most costly defects are HFN and LFN. Based on this more informative data, it is better todevelop an improvement project to reduce HFN and LFN type of errors.

N.B. Here data up to 31stMay 2012 is taken into consideration.

Table 4 - Confidence Interval for One-Sample Poisson Rate

Length of observation = 1, * Data up to 31st May 2012

Total number of errors

Variable

7

7

N

0.8961

Rate of occurrence

(0.6972, 1.134)

95 % confidence interval

69*

Total occurrences



tistic is central to the capability study,without sufficient data to estimate meanDPU, the analysis can not continue.

3) Poisson plot - Poisson plot is locatedin the upper right corner of the processcapability analysis. Poisson plot plotsthe expected and observed number oferrors. It can be seen from the graphthat the plotted points are in the straightline. This proves that the assumptionthat the data were sampled from a Pois-son distribution is correct.

4) Histogram of distribution of DPU - Thisis located in the lower right corner offig. 5. The distribution of DPU graphshows that the errors per RBRC followPoisson distribution. This graph showsa vertical line for the arbitrary target offinding 2 errors per session of RBRC.

Fig. 5 - Poisson capability analysis of RBRC sessions

5) Summary statistic table - It is located inthe lower center of the process capa-bility analysis. It consist of the follow-ing

a. Mean DPU - An estimate of theaverage number of errors perRBRC (0.8961) as well as con-fidence intervals for the esti-mates (0.6972- 1.1341).

b. Minimum and Maximum DPU -Range is between 0.00 to 4.00

c. Target DPU - The target numberof errors specified by theprogramme managers. Here thevalue is 2 errors per RBRC ses-sion.

Discussion:

On site evaluation and panel testing failed todetect any error in examination of AFB slides

Swapnil Vishnu Lale

41



in Solapur district. These observations differfrom those reported by Selvakumar N, et alwhile studying panel testing in Chennai, SouthIndia. They have reported 15 errors out of 360slides examined in panel testing [5]. This indi-cates that when the Panel of slides was exam-ined carefully in presence of STDC, Pune teamSTLSs perform less errors. But Panel testingdoes not check routine performance. This pointtowards negligence or casual attitude at rou-tine work and not to technical incompetence.Since January 2006, 42191 slides were re-checked. Different types of 69 errors weredetected in RBRC. Similar findings were re-ported by Malik S et al; while studying RBRCprogramme in Delhi [6]. Most common typeof error has been quantification error followedby high false negatives in the present study.These findings differ from those observed byYip C.W. et al in a study at Hong Kong. Theyhave reported LFN and LFP as most commonforms of errors. This difference may be due tothe florescence microscopy procedure they uti-lized [7] . In a Mexican study; Martinez A et alreported that RBRC detects large number oferrors, mostly quantification errors [8]. Thisstudy has found out that the distribution of er-rors is significantly different amongst varioustuberculosis units. TU and DMC wise differ-ences are reported in NRHM document of highfocus state of Jharkhand [9]. In this study; aver-age DPU has been found to be 0.8961 with arange from 0 to 4 and confidence interval of0.697 to 1.134. Similar observations have beenmade by Bais R while using capability index toimprove laboratory analytical performance. Butthey have broadened the range of variability indetection of errors up to 6 δ (standard devia-

tions) on each side of the mean (µ) [10].Limitations of study:

This study has several limitations. Inherentstructure of lot quality assurance sampling doesnot permit to make precise estimates of thepopulation parameters. So the results of thisstudy may not be generalizable. Certain biasesarise because the investigator is a part of theprogramme management team. For the retro-spective part of the study, investigator is un-aware of the micro-details of the ground levelsituation. However he may think of them fromhis present experience. A precise estimate ofthe cost of consequences of un-treated TB pa-tients is not available for Indian settings. There-fore Asian values are utilized [4]. They may notbe appropriate in local settings. The strengthsof this study include the use of various qualityassessment tools in Minitab software. Thesoftware is established and widely applied [11].Using secondary data entails a massive amountof cost and time savings while permitting largesample size.

Conclusions and Recommendations:

More number of errors have been found during2006-07 period. Incidentally panel testing hasnot been performed during that period. This find-ing stresses that panel testing should be con-ducted regularly. TU wise analysis shows thatBarshi and Mangalwedha TU have got morenumber of errors than Karmala TU. When sub-jected to statistical analysis, the relationshipof TU with number of errors is statistically sig-nificant. So programme manager should paymore attention to poorly performing units likeBarshi and Mangalwedha TU. Concerned labo-ratory technician may be trained, infrastructure

Swapnil Vishnu Lale

42



of the laboratory may be revamped, and vacantpositions need to be filled up.Pareto chart shows more number of quantifi-cation errors. Weighted Pareto chart takes intoaccount the cost of errors. High false negativesand low false negatives turn out to be the cost-liest forms of errors. As a consequence of falsenegative error, patients with TB may not betreated, resulting in suffering, spread of TB anddeath. Intensive phase of treatment may not beextended for the required duration and as a re-sult patient may loose confidence in theprogramme.At the time of collection of sputum it is im-portant to make sure that the sample containssputum, not just saliva, that too in adequatequantity. (at least 2 ml) The Laboratory techni-cian should select thick, purulent particles tomake smear and prepare smears correctly- nottoo thick, too thin or too little material. Slidesshould be fixed for correct length of time andstained with Carbol fuchsin for the full fiveminutes. De-colourization with sulphuric acidshould not be done too intensively. Smearsshould be examined for at least five minutesbefore recording it as negative. They shouldmake sure to label the sputum containers, slidesand laboratory forms carefully. Last but not theleast cross check the number on the laboratoryform and sputum container before recording.The number of errors detected in each RBRCsession seems to follow a Poisson distribution,with rate of occurrence equal to 0.8961. Theprocess of RBRC is in control with fewer num-ber of points located outside the control lim-its. Cumulative DPU rate stabilizes after 55th

RBRC session onwards at about 0.89. The as-sumption of Poisson distribution is found cor-

rect and the target of 2 DPU per RBRC sessionis feasible. The process of RBRC is capable toachieve desired target of detection of errors.

References:

1. Tuberculosis Control in India17.pdf[Internet]. [cited 2012 Oct 7].h t t p : / / t b c i n d i a . n i c . i n / p d f s /Tuberculosis%20Control%20in%20India17.pdf

2. Solapur District Population Census 2011,Maharashtra literacy sex ratio and density[Internet]. [cited 2012 Oct 7]. http://www.census2011.co.in/census/district/364-Solapur.html

3. Managing the revised national tuberculosiscontrol programme in your area, A trainingmodule. First ed. New Delhi: Central TB di-vision, Ministry of health andFamily welfare;2006.

4. Baltussen R. Cost effectiveness analysis ofstrategies for tuberculosis control indeveloping countries. BMJ 2005; 331(7529):1364.

5. Selvakumar N, Sivagamasundari S,Prabhakaran E, Govindaraju R, Perumal M,Granich R, Wares F, Chauhan LS, NarayananPR. Storage of heat-fixed unstained sputumAFB smears for panel testing in a tuberculo-sis unit in South India. Int J Tuberc Lung Dis2005; 9(2):223�5.

6. Malik S, Hanif M, Chopra KK, Aggarwal N,Vashist RP. Evaluation of a new quality as-sessment strategy for blinded rechecking ofrandom sputum smears for TB in Delhi, In-dia. Southeast Asian J Trop Med PublicHealth 2011; 42(2):342�6.

7. Yip CW, Chan MY, Cheung WF, Yu KW, TangHS, Kam KM. Random blinded rechecking

Swapnil Vishnu Lale

43



of sputum acid-fast bacilli smear using fluo-rescence microscopy: 8 years� experience.Int J Tuberc Lung Dis 2012; 16(3):398�401.

8. Martinez A, Balandrano S, Parissi A, ZunigaA, Sanchez M, Ridderhof J, Lipman HB,Madison B.. Evaluation of new external qual-ity assessment guidelines involving randomblinded rechecking of acid-fast bacillismears in a pilot project setting in Mexico.Int J Tuberc Lung Dis 2005; 9(3):301�5.

9. Jharkhand_Report .pdf [Internet]. [cited2012 Oct 7].

* Author for Correspondence: Dr. Swapnil Vishnu Lale, District Tuberculosis Officer, Civil hospital,Solapur - 413002, (Maharashtra), India. Cell No. - 9421958419, E mail- [email protected]

http://www.mohfw.nic.in/NRHM/Docu-m e n t s / H i g h _ F o c u s _ R e p o r t s /Jharkhand_Report%20.pdf

10.Bais R. Use of Capability Index to ImproveLaboratory Analytical Performance. ClinBiochem Rev 2008; 29(Supplement(i)):S27�S31.

11. Henderson GR. Six Sigma Quality Improve-ment with Minitab. John Wiley and Sons;2011.

Swapnil Vishnu Lale

44

evaluation of random blinded re-checking of afb slides under

Documents