Surabhan Ratanasen, a doctor who filed for bankruptcy under Chapter 11, appeals the district court’s order affirming the bankruptcy court’s finding that Ratanasen owed the California Department of Health Services $125,789 under the Medi-Cal program based on an audit conducted through sampling and extrapolation. We find the use of such audit methods proper and that Ratanasen was afforded an opportunity to rebut the audit results, and we affirm the district court.

FACTS AND PROCEEDINGS BELOW

Ratanasen, a physician providing services under the Medi-Cal program, was the subject of an investigation by the California Department of Justice (“DOJ”) Medi-Cal fraud unit between October, 1984 and March, 1987. About 275 patient files were seized from the doctor’s office in November, 1986; he later voluntarily released a number of additional files to investigators.

Felony charges of presenting false claims in violation of California Welfare and Institutions Code section 14107 were filed against Ratanasen in May, 1987.

In December of 1986, the Department of Health Services (“DHS”) audited claims for payments filed by Ratanasen between May, 1984 and March, 1986. Prior to the final audit report, two exit conferences were scheduled with Ratanasen to explain the preliminary findings of the medical review team. Appellees claim that Ratanasen canceled both conferences. Ratanasen claims that in light of the pending criminal charges against him, he requested that either the exit interview be continued until after the criminal proceedings had been resolved, or the attorney general’s office stipulate that statements he made at the conference would not be used against him. Ratanasen claims the DOJ denied his request.

In May 1988, Ratanasen pled guilty to four counts of a lesser offense of violating Business and Professions Code section 2261. He stipulated that the basis for his plea was that he falsified patients’ medical records and submitted payment claims to Medi-Cal for services he had not actually rendered. He filed a Chapter 11 petition for bankruptcy in July, 1988.

A final DHS audit of the records released in October, 1988, determined that Ratanasen had overbilled the Medi-Cal program $124,-268, and that future claims would have to be supported by copies of patient records when submitted for payment. The audit was conducted by selecting a sample of 300 Medi-Cal beneficiaries out of a total of 8,761 beneficiaries for whom Ratanasen had submitted claims during the period in question. The doctor failed to appeal the audit findings and they became final.

Ratanasen filed an objection to the allowance of the claim in bankruptcy court in July, 1989, alleging that the claim was the result of a random sample, which was then used to calculate an estimated overpayment. The objection further alleged that to reach a true overpayment, each file would have to be examined on its own. The bankruptcy court, on its own motion, bifurcated the legal issue of whether or not statistical reports are allowable to prove the existence and amount of a claim in bankruptcy court. It concluded that a creditor may prove the amount of its claim through the use of convincing statistical samplings. The bankruptcy court then held an evidentiary hearing and concluded that the method of statistical extrapolation used by the DHS was valid and in compliance with California law. It further found that the fact that Ratanasen did not receive an exit conference was of his own choosing and thus a hearing was not required in order for the bankruptcy court to enter judgment.

Judgment in the amount of $124,268 was entered for the state in February, 1991. Ra-tanasen appealed to the district court, which affirmed the decision of the bankruptcy court, and then timely appealed to this court. We have jurisdiction pursuant to 28 U.S.C. § 158 (d) (Supp.1992).

DISCUSSION

1. Standard of review.

The district court’s decision is reviewed de novo. In re Ewell, 958 F.2d 276, 279 (9th Cir.1992). “The factual findings of the bankruptcy court are reviewed for clear error, and its conclusions of law are reviewed de novo.” Id.

Following the evidentiary hearing, the bankruptcy court issued a memorandum opinion, which contained the following statements under the heading, “Findings of Fact:”

Ratanasen maintains that No. 9 contains conclusions of law, or at least involves mixed issues of law and fact. His latter assertion is correct. As explained further below, whether the use of sampling and extrapolation is. proper is a question of law, while whether the sample size, etc., were appropriate is a question of fact.

He further alleges that No. 12 involves mixed issues of law and fact and ■ as such merits de novo review. We determine that whether or not Ratanasen had the opportunity to rebut the state’s findings is solely a question of law.

A conclusion of law cannot be turned into a finding of fact simply by labeling it as such. In re Bubble Up Delaware, Inc., 684 F.2d 1259, 1262 (9th Cir.1982). However, because we review the bankruptcy court’s mixed findings, and the district court’s decision, de novo, any error that may have occurred from the bankruptcy court’s inaccurate classification of its determinations, or the district court’s application of the wrong standard of review, is “completely harmless.” In re Marquam Inv. Corp., 942 F.2d 1462, 1464 (9th Cir.1991).

2. Validity of using sampling methods.

Whether or not the DHS could prove its claim by using statistical information is a matter of law. The use of sampling and extrapolation as part of audits to determine overpayments to parties who receive publicly-funded reimbursements has not been addressed by this circuit. A Georgia district court addressed this issue in State of Georgia Dept. of Human Resources v. Califano, 446 F.Supp. 404 (N.D.Ga.1977), which involved the Department of Health, Education and Welfare’s (“HEW”) refusal to reimburse Georgia for $3.5 million the state paid to doctors who provided services to Georgia Medicaid recipients for three years in the 1970s. An audit, conducted on the basis of random statistical samples of claims paid during a five-quarter period, revealed that Georgia had paid some claims in excess of ceilings imposed by federal statutes. Id. at 406 . As a result, HEWs Social Rehabilitation Services (“SRS”) regional commissioner disallowed $2.8 million in matching federal funds and made a demand for a refund of that money. Following an appeal to the SRS administrator, some claims were found allowable, but a disallowance of $1.5 million was upheld. Id. Georgia appealed, claiming that the administrator’s decision was arbitrary and capricious because the amount of overpayment was determined by use of a statistical sample rather than by an individual, claim-by-claim.review. In finding for defendant HEW, the district court concluded:

Id. at 409 (citations omitted).

The Seventh Circuit cited Georgia in upholding the auditing procedures used by the state of Illinois in auditing physicians who are reimbursed with public funds for medical services. Illinois Physicians Union v. Miller, 675 F.2d 151, 156 (7th Cir.1982). The state audited a sample of 353 records randomly selected from a total of 1,302 records for the audit period, and determined a participating doctor had been overpaid $5,018. Id. at 152 . Plaintiff doctor contended any formula for sampling and extrapolation is improper per se. Id. at 155 . The court disagreed and concluded that “the use of sampling and extrapolation is proper provided there is an opportunity to rebut the initial determination of overpayment.” Id. at 156 .

The Sixth Circuit also cited Georgia in a case where the Department of Education found, as a result of an audit conducted through random, stratified sampling, that the Michigan Department of Education misspent federal funds in the conduct of its vocational rehabilitation program. Michigan Dept. of Educ. v. United States Dept. of Educ., 875 F.2d 1196 (6th Cir.1989). The Michigan court found, as in Georgia, that audits of thousands of cases “comprising the universe of cases” would be impossible. Id. at 1205 . Additionally, as in Georgia, the final determination of the invalid expenditures was not made until the state had a chance to present its own evidence of an error in the audit. Id.

Two other circuits approved of the use of statistical sampling in 1991 cases. Chaves County Home Health Serv., Inc. v. Sullivan, 931 F.2d 914 (D.C.Cir.1991), cert. denied, — U.S. -, 112 S.Ct. 1160 , 117 L.Ed.2d 408 (1992), involved home health care providers who contended that the Secretary of Health and Human Services (“HHS”) improperly suspended the existing individual claims adjudication process under the Medicare Act and replaced it with a method based on statistical sampling to determine overpay-ments. The District of Columbia Circuit cited to Georgia, Illinois and Michigan in affirming summary judgment for HHS, saying:

Id. at 919.

In Yorktown Medical Lab., Inc. v. Perales, 948 F.2d 84 (2d Cir.1991), a Medicaid provider claimed that the statistical sampling techniques used by the state for calculating overcharges in the provider’s claims violated its due process rights. The district court found that the provider had failed to establish that it had a cognizable property interest and granted summary judgment for defendant. In affirming the district court, the Second Circuit explained that the provider’s claim overlooked the fact that the due process required depends on balancing various circumstances and factors as set out in Mathews v. Eldridge, 424 U.S. 319 , 96 S.Ct. 893 , 47 L.Ed.2d 18 (1976). Citing to Chaves and Illinois, the court found, “Given the low risk of error and the government interest in minimizing administrative burdens, the balance of interests favors [the state].” Yorktown, 948 F.2d at 90.

We now join other circuits in approving the use of sampling and extrapolation as part of audits in connection with Medicare and other similar programs, provided the aggrieved party has an opportunity to rebut such evidence. To deny public agencies the use of statistical and mathematical audit methods would be to deny them an effective means of detecting abuses in the use of public funds. Public officials are responsible for overseeing the expenditure of our increasingly scarce public resources and we must give them appropriate tools to carry out that charge.

3. Were the sample size, level, etc. appropriate?

Defendant has cleared the initial legal hurdle with our decision that an agency may use statistical information to prove a. claim . The next question is whether or not the bankruptcy court erred in finding that the sample size, level and other measures of the extrapolation method were appropriate and convincing. If they were, then the final question for the court is whether or not the aggrieved party had an opportunity to rebut the claim.

Appellant points out that in this case, the DHS used a simple random sampling technique to create the sample population. Appellant contends that simple random sampling is an appropriate way to create a sample population from a homogenous population, but that it is not appropriate in a situation such as this one involving a heteroge-nous population, where stratified random sampling should be used. For this proposition. he cites to Grier v. Kizer, 219 Cal.App.3d 422 , 268 Cal.Rptr. 244 (1990). However, in that case, the California Court of Appeals ruled that the challenged audit method should have been adopted pursuant to the state’s Administrative Procedure Act; failure to so adopt rendered the audit method invalid and unenforceable. The court declined to reach the statistical validity of the method. Id. 268 Cal.Rptr. at 256 .

Appellant also contends that the sample size of 3.4 percent of the population was so small as to represent a violation of due process. He cites to Daytona Beach General Hospital, Inc. v. Weinberger, 435 F.Supp. 891 (M.D.Fla.1977), a case in which a recoupment due from a hospital because of alleged overpayments was calculated by using a sampling method based on less than ten percent of the total cases in question. The plaintiff hospital never received a hearing prior to a final determination of any overpayments and had no access to an appeals procedure. The district court found that the ten percent procedure used denied the plaintiff due process and remanded the case for the Secretary of HEW to review, a review which the court said should include an opportunity for the plaintiff to be heard. Id. at 900 .

Unlike the hospital in Daytona, appellant in this case did have access to hearings and appeals and was afforded an opportunity to be heard. Additionally, the cases addressing sampling cited above make no mention of a statistical “floor” which auditors must exceed in order to guarantee providers due process. Indeed, the sample of 3.4 percent in the instant case exceeds that of the sample in Michigan, where a random, stratified sample of .4 percent was used as a starting point for determining improper expenditures. Michigan, 875 F.2d at 1199 .

The bankruptcy court heard from three witnesses during an evidentiary hearing on Ratanasen’s objections to the state’s claim. One of those witnesses was Fan Yee, a research program specialist for the state who testified that the simple random sampling method chosen for this case- was “appropriate, valid and reliable.” Ratanasen’s expert, Dr. Michael Intriligator, testified that the appropriate method to sample would have been stratified random sampling, and that the state’s method was invalid and unreliable. The bankruptcy judge found that Intriligator was not as convincing as Yee, and found that simple random sampling was a valid method to use, as described in “Findings of Fact” No. 9.

We conclude that the bankruptcy court and district court did not err in upholding the validity of the sampling methods.

4. Did Ratanasen have adequate opportunity for an exit hearing?

Appellant challenges the bankruptcy judge’s findings of fact that he did not receive an exit conference by his own choice and that such an interview was not required in order for the bankruptcy court to enter final judgment. He claims that whether or not he had an adequate opportunity for an exit hearing is a mixed question of law and fact, and as such merits de novo review.

As noted above, courts have approved the use of sampling and extrapolation on the condition that the party charged with over-payments has the opportunity to rebut the initial determination of overpayments. Additionally, California law requires that a medical services provider “shall be afforded a reasonable opportunity to participate in an exit conference after the conclusion of any field audit ... and prior to the issuance of the Audit Report,” at which time the provider can present relevant information concerning the audit findings. Cal.Code Regs. tit. 22, § 51021(a). We find that the question of whether or not Ratanasen had an adequate opportunity for an exit interview is a question of law and as such review the bankruptcy court’s decision de novo.

The exit interview would have afforded Ratanasen the opportunity to rebut the initial determination of overpayments. Thus, if he was provided with the opportunity for an exit interview, due process concerns about the use of sampling and extrapolation would be addressed.

Appellant claims that he was effectively denied an opportunity for a hearing by being forced to choose between a hearing on one hand and his constitutional right against self-incrimination on the other.

The state points to the record to establish the following chronology:. (1) an exit conference was scheduled on Nov. 20, 1987; (2) it was rescheduled at Ratanasen’s request for Feb. 1, 1988; (3) Ratanasen’s lawyer canceled the Feb. 1, 1988 conference; (4) a request made by Ratanasen on March 11, 1988 to reset the conference was denied and he was notified of his appeal rights.

As to whether or not Ratanasen’s rights were adequately protected:

Illinois, 675 F.2d at 157 (citing Mathews v. Eldridge, 424 U.S. 319, 335 , 96 S.Ct. 893, 903 , 47 L.Ed.2d 18 (1976))

We find the state provided reasonable opportunity for him to challenge the initial audit results. Ratanasen sought to delay the hearing and by his own choice canceled two hearings that had been scheduled by the state for his own benefit because he was under a criminal investigation regarding these same matters; thus he was not prejudiced by the state’s decision to proceed with the audit. Ratanasen did not appeal the audit determination, which then became final.

To require the state to provide an extraordinary opportunity for Ratanasen to respond would be an administrative and financial burden not warranted by the situation presented here. Also significant is the fact that the bankruptcy court held an evidentiary hearing on Ratanasen’s objection to the state’s claim, at which time Ratanasen had an opportunity to fully air his objections to audit methods.

We find that Ratanasen’s due process rights were not violated.

5. The money judgment against Ratana-sen.

Ratanasen claims that the bankruptcy court exceeded its authority by entering judgment against the debtor following its memorandum opinion concluding that the state had proven its claim. He further alleges that there is no statutory authority for entering judgment against a debtor-in-possession on a prepetition claim. The state maintains that Ratanasen may not raise this issue for the first time on appeal, and because he did not raise this issue in the district court, it is deemed waived.

Diamond Door Co. v. Lane-Stanton Lumber Co., 505 F.2d 1199, 1206 (9th Cir.1974) (footnotes omitted). The exceptions do not apply in this case. Ratanasen has waived this issue on appeal.

CONCLUSION

The decision of the district court is AFFIRMED.

. The cases discussed above do not specify that a certain method of sampling must be used to satisfy due process. In Michigan, the auditors did apparently use a "random, stratified sample,” Michigan, 875 F.2d at 1199 , but in discussing the issue the court refers only in general terms to the "validity of the use of random sampling.” Michigan, 875 F.2d at 1205 . In the other aforementioned cases the samples were “randomly selected,” Illinois, 675 F.2d at 152 ; a "random statistical” sample, Georgia, 446 F.Supp. at 406 ; a "statistical sampling,” Chaves, 931 F.2d at 915 ; and "statistical” and "random,” Yorktown, 948 F.2d at 86. . The Grier court explained the difference: “An illustration given is of a sample drawn from a box containing 500 pounds of oranges and diamonds. A random 10 pound sample might yield a misleading value if it contained a disproportionate number of diamonds. Stratified sampling, as contrasted with random sampling, would draw separate samples of diamonds and oranges.” Grier, 268 Cal.Rptr. at 247 n. 3. . Grier involved the same two experts on statistical sampling who testified in this case. In both cases, Fan Yee appeared for the state and testified as to the accuracy to the random sampling method, while Michael Intriligator, the doctors' expert, testified that a stratified sampling method would have yielded a more accurate result. Grier, 268 Cal.Rptr. at 247 . . The Michigan court noted, "There is no case law that states how large a percentage of the entire universe must be sampled.” Michigan, 875 F.2d at 1206 .